Understand your school’s AI safeguarding and governance position
AISG is a structured self-review for schools in England. Across 58 controls — short, checkable statements about your arrangements — it helps you establish what can be evidenced, where gaps remain and what may need attention.
2–3 hours · Six domains · Saved on your device · No account required
Practitioner-developed pilot methodology. Not an inspection or certification.
Nothing leaves your device. No account · no server upload · no analytics · no external fonts. Your review is stored in this browser and exported only when you choose.
Leadership
Safeguarding
Data protection
Technology
Teaching & staff practice
Assurance & improvement
Before the review
Start with the one-page briefing for your role
Three one-page briefings explaining the principal issues for each audience. The eleven documents below are produced later, from your completed return.
Staff
Using AI safely at school
What staff need to know about using AI, protecting information, checking outputs and reporting safeguarding concerns.
Senior leaders
Leading AI safely and responsibly
What leaders need to know about AI use, safeguarding, data protection, approving tools and providing governance assurance.
Governors
What boards need to know about AI
The questions governors should ask, the evidence they should expect and the significant risks they should oversee.
Staff · Editable safeguarding example
Nude or sexual image of a pupil: Report Remove staff briefing
An editable one-page quick reference for staff: what to do and not do, the four steps, special cases, how a pupil uses the Childline and IWF Report Remove tool, and when to involve the DSL. Add your school’s own contacts, then print or save.
One completed return can produce eleven role-specific records, briefings and statements. The three most significant:
01Governing board briefing — the position, the significant gaps, and what the board should ask next
02Leadership brief — the six responsibilities, what good evidence looks like, and the questions leaders should be able to answer
03Full evidence record — all 58 controls with ratings, evidence references, referrals and the action plan
View all 11 outputs
04Safeguarding practice extract — for the designated safeguarding lead
05Data protection extract — for the data protection officer
06Technology and filtering extract — for the IT lead and the external provider
07Staff statement — what staff are told, conditional on the evidence recorded
08Parent and carer statement — a plain-English account for families, conditional on the evidence recorded
09Pupil statement — age-appropriate wording, conditional on the evidence recorded
10Trust or diocese summary — a standard return for comparison and moderation
11Alignment note — to sit alongside an annual safeguarding self-evaluation
Evidence-aware reporting. In the staff, parent and pupil statements, a sentence saying something is in place appears only where the control behind it is fully evidenced; where it is not, the wording says the school is reviewing that area. Every document states that it comes from a and has not been independently checked. A separate packages the whole return for independent review.
Governance that can keep pace with change
AI tools will change, but the responsibilities schools hold do not disappear when the technology changes. AISG focuses on the institutional controls surrounding AI use rather than assessing individual AI products — and identifies where action or specialist review may be needed.
Evidence-ledReview what can actually be demonstrated, not simply what a policy says.
TraceableEvery control is connected to an underlying source or stated methodological rationale.
RepeatableReturn to the review as systems, evidence and guidance change.
TransparentMethodology versions, evidence gaps and limitations remain visible.
Developed from school leadership practice
AISG grew from a practical question encountered in school leadership: how can a school demonstrate that its arrangements for AI safeguarding and governance operate in practice, rather than simply having policies in place? The initial case study documents its application in one school setting, including what the review identified and what was learned.
·
Pilot edition · Autumn term 2026
During the Autumn 2026 pilot, participating schools are being asked to complete the review, record the time it takes and provide structured feedback on clarity, usability and evidence requirements. Findings will inform the next methodology revision. Pilot use is not validation.
To join, write to w.gopaul@st-annes.bham.sch.uk with your school name, your role, and the term you would complete the return in.
· Application 0.5.8 · ·
AISG — AI Safeguarding & Governance
Ready to review your school’s arrangements?
A structured self-review across the six domains above. Your work is saved on this device as you go; you can stop and resume at any point.
The review helps you:
✓Record evidence
✓Identify gaps
✓Assign actions
✓Report to governors
Prints as an A4 document. Choose “Save as PDF” for a copy, and set Headers and footers to off in the print dialog so no browser URL or date appears on the page.
AISGAI Safeguarding & Governance
Staff briefing
Using AI at school
What every member of staff needs to know before using AI, and what to do when something looks wrong.
Published
August 2026
Prepared against
KCSIE 2026, which comes into force 1 September 2026
Briefing ID
AISG-BR-STAFF-2026-08-16
Version
0.5.8
These expectations apply to every member of staff, whatever your role. AI tools can save time, but used carelessly they put pupils’ information, safeguarding and professional judgement at risk.
1 · CheckIs this an approved tool, for this use?
2 · ProtectEnter nothing sensitive or identifying.
3 · VerifyCheck the output before it goes anywhere.
4 · EscalateReport concerns to the DSL.
Before using an AI tool
Use for approved purposes. Approval applies to both the tool and the intended use.
If a tool or use has not been approved, check first. The approval route applies to free tools too.
Never enter a child’s name, personal details or anything identifying into an AI tool that has not been approved for that information.
When AI helps you produce work
If AI helps you produce anything that affects a pupil, read and check it before it goes anywhere.
writes fluently and is sometimes wrong. Treat what it gives you as a draft. You remain responsible for the final output.
Safeguarding
An AI-made or altered image of a child is a safeguarding matter and can be a criminal offence. Report it to the like any other concern.
AI never replaces safeguarding judgement or established reporting routes.
Pupils and AI
Pupils should use AI for their work only where a teacher has permitted it, and should say when they have.
Help pupils understand that AI output needs checking, and that making or sharing fake images of real people is never a joke.
If you are unsure
Check before you act. Speak to a senior leader about tools and uses, and to the DSL about anything with a safeguarding dimension. You do not need to be certain before you raise a concern.
Stop and ask first when it involves safeguarding information · a pupil’s personal data · a significant decision about a person · a tool you are not sure is approved · content that could be harmful.
If you are the DSL: when the school completes an AISG Evidence Review, a safeguarding practice extract of the full return is produced for you.
Key terms
1 · Approved tool
A tool the school has agreed can be used, for a stated purpose. Approval covers both the tool and the intended use, and applies to free tools too.
2 · Generative AI
AI systems that produce new text, images, audio or video in response to a prompt — for example, chat assistants.
3 · DSL
Designated Safeguarding Lead: the member of staff responsible for safeguarding and child protection, including online safety and AI-related concerns.
Approved tool · Approved use · Protect information · Check the output · Follow safeguarding procedures
Prints as an A4 document. Choose “Save as PDF” for a copy, and set Headers and footers to off in the print dialog so no browser URL or date appears on the page.
AISGAI Safeguarding & Governance
Leadership briefing
AI governance: leadership brief
The six responsibilities the senior leadership team holds for AI, and the evidence that shows they are working.
Published
August 2026
Prepared against
KCSIE 2026, which comes into force 1 September 2026
Briefing ID
AISG-BR-SLT-2026-08-16
Version
0.5.8
The governance cycle
Repeats annually
1Baseline annual reviewEstablish the evidence position
2Action planOwners and target dates
3Termly oversightProgress to leaders and governors
4Annual reassessmentReview again and compare
The cycle repeats each year. Material changes to an AI product, its functionality, terms or data handling may require proportionate re-review.
AI governance exists so the school knows what AI is in use, makes proportionate decisions about the risks, and can show that safeguarding, privacy and professional accountability still hold. It is not there to stop useful innovation.
Six leadership responsibilities
Know the
Keep a credible picture of approved and material unapproved AI use, including free, embedded and browser-based tools.
Set clear boundaries
Staff and pupils should know which tools and uses are permitted, what must never be entered into an unapproved system, and where concerns go. Approval applies to both the tool and the intended use.
Protect safeguarding practice
Existing procedures must be capable of responding to AI-enabled harm. AI must not replace DSL judgement, professional safeguarding decision-making or established safeguarding routes.
Control adoption
New AI systems go through proportionate review: safeguarding, privacy, age appropriateness, data handling, contract and vendor terms, and technology risk where relevant. Approval is not permanent.
Keep humans accountable
Staff remain responsible for professional outputs and for significant decisions about pupils. AI-generated information is checked and, where needed, challenged.
Assure the governing board
Give governors evidence, not reassurance alone. Say what is working, what is incomplete and what action is under way.
What good evidence looks like
Evidence of operation matters, not policy existence alone. An up-to-date AI inventory. Completed tool reviews. Privacy and decisions, including documented decisions that a DPIA was not needed. Staff guidance and training records. Safeguarding procedures that would manage an AI-related incident. review records. Governor oversight. Tracked actions with owners and dates.
Six questions leaders should be able to answer
What are we using?
Why are we using it?
What are the risks?
Who owns those risks and how are they managed?
What evidence tells us the controls are working?
What happens when something changes or goes wrong?
Key terms
1 · AI estate
The full picture of AI in use across the school: approved tools, AI embedded in other products, and any material unapproved use.
2 · DPIA
Data Protection Impact Assessment: a structured assessment used to identify and reduce data-protection risks before personal data is processed. Whether one is needed is a judgement for the data protection lead or DPO.
3 · Filtering and monitoring
The systems that block unsuitable online content and flag concerning activity on school devices and networks. KCSIE expects them to be reviewed at least annually.
The Evidence Review is a point-in-time practitioner methodology. It does not certify compliance, guarantee safety, or replace safeguarding, data-protection or other school policies. Self-assessment is not independent assurance.
Prints as an A4 document. Choose “Save as PDF” for a copy, and set Headers and footers to off in the print dialog so no browser URL or date appears on the page.
AISGAI Safeguarding & Governance
Governor briefing
AI governance: what boards need to know
What a governing board should expect to be told about AI, and how to tell evidence from reassurance.
Published
August 2026
Prepared against
KCSIE 2026, which comes into force 1 September 2026
Briefing ID
AISG-BR-GOV-2026-08-16
Version
0.5.8
Governors do not need to become AI specialists. They do need appropriate assurance that safeguarding, data protection, technology, teaching and governance arrangements remain effective as AI use develops.
1 · AskWhat is being used, and how do leaders know?
2 · EvidenceWhat shows the controls operate?
3 · ChallengeWhere are the gaps?
4 · AssureWho owns the actions, and when does progress come back?
Six areas of assurance
Leadership & accountability
A named senior leader coordinates AI governance, and the board receives considered assurance rather than passing mentions.
Safeguarding & pupil welfare
Safeguarding procedures can respond to AI-enabled harm, and AI-related concerns follow established reporting routes.
Data protection & privacy
Leaders know which tools handle pupil information, and privacy decisions are documented, including decisions that a was not needed.
Technology & procurement
New tools are reviewed before use with pupils or pupil data, and keep pace with AI services.
Teaching & staff practice
Staff know what is permitted, check AI-assisted work, and remain accountable for professional judgements.
Assurance & improvement
Arrangements are reviewed on a cycle, actions are owned and tracked, and evidence shows the controls operate.
The board’s role
Oversight: assurance, challenge, and confidence that risks are owned and progress is reported.
Leadership’s role
Operation: choosing tools, running the controls and managing incidents day to day. The board does not manage systems.
Five questions governors should ask
What AI systems are currently being used by staff and pupils, and how do leaders know?
What happens before a new AI tool can be used with pupils or pupil information?
Could our safeguarding procedures deal effectively with an AI-enabled safeguarding concern today?
How do leaders know filtering, monitoring, data-protection and staff-training arrangements remain effective?
What are the most significant gaps identified by the school’s latest review, who owns them and when will they be addressed?
When the school completes an AISG return, the governing board briefing it produces restates these as school-specific questions, with the evidence position beside them.
How ratings are decided
Ratings rest on evidence a school can produce, not on intent. When in doubt the claim is downgraded, not raised.
What good assurance looks like
Governors should distinguish between evidence that a policy exists and evidence that arrangements operate in practice. A policy on file says what should happen; minutes, review records, training logs and completed checks show what does happen.
Where the school has completed its own review, ask what was found, what actions follow, who owns them and when the board will hear about progress. A is not independent assurance.
Key terms
1 · DPIA
Data Protection Impact Assessment: a structured assessment used to identify and reduce data-protection risks before personal data is processed.
2 · Filtering and monitoring
The systems that block unsuitable online content and flag concerning activity on school devices and networks. KCSIE expects them to be reviewed at least annually.
3 · Self-assessment (Tier 1)
A review the school completes itself. Evidence is declared or internally reviewed; nobody outside the school has checked it.
Who built AISG, why it exists, what it does not claim to be, and how to get in touch.
Published
August 2026
Prepared against
KCSIE 2026, which comes into force 1 September 2026
Document ID
AISG-DOC-ABOUT-2026-08
Version
0.5.8
Why AISG exists
Schools were being asked to account for AI use before anyone had told them what accounting for it looked like. KCSIE 2026 made the safeguarding expectations explicit, and governing boards started asking questions that leaders could not yet answer with evidence rather than intent.
The gap was not policy. Most schools could produce a policy. The gap was being able to show that the arrangements described in it actually operate, and knowing which parts had never been checked. AISG is the structure used to work that out in one school, made general enough for others to use.
Built from school leadership practice
AISG began with a practical question: how can a school demonstrate that it is governing AI responsibly, rather than simply having an AI policy?
The methodology has been developed through the founder’s work as a serving headteacher and through its application within a real school setting, which provides the initial practitioner case study. School leadership provides the operational context for the governance, risk and accountability questions AI creates.
Bringing school leadership and AI governance together
School leadership experience
AI governance focus
Strategic and operational leadership
Responsible AI governance
Safeguarding responsibility
AI risk and harm
School governance and accountability
Governance frameworks and accountability
Policy implementation
AI lifecycle governance
Evidence and assurance
Risk management and monitoring
Data protection responsibilities
Privacy and data governance
Staff development
AI literacy and responsible implementation
Working with governors and external professionals
Oversight, challenge and assurance
This combination shapes the approach AISG takes: AI governance is treated as an institutional responsibility involving safeguarding, leadership, evidence, privacy, security and professional judgement — not simply as a technology issue.
About the founder
Wendell Gopaul Headteacher · Founder, AISG
A serving headteacher in Birmingham, he has been headteacher and designated safeguarding lead at St Anne’s Catholic Primary School since 2016, a maintained primary school in an area of high deprivation. He has also worked as a school inspector.
Alongside headship he designs and delivers professional development for teachers and school leaders, including training on the safe and accountable use of AI in schools, and advises on education strategy internationally.
AISG came out of that work. The tool is the method he uses in his own school, written down so other schools can use it, and published with its sources, rules and limitations visible so that others can scrutinise it. His experience explains where the methodology comes from; it is not offered as evidence that the methodology has been independently validated.
What AISG is not
It is not a certification, a legal opinion, a technical security audit, a statutory inspection or a validated assessment instrument. It does not certify compliance, guarantee safety, or replace safeguarding, data-protection or other school policies. A self-assessment is not independent assurance.
Ratings rest on evidence a school can produce, not on intent. When in doubt the claim is downgraded, not raised.
Who AISG is for
School leaders — the overall AI governance position and its priorities. Designated safeguarding leads — where AI use intersects with safeguarding arrangements and where further review may be needed. Governors and trustees — oversight, challenge and accountability. DPOs and privacy professionals — the AI-related areas requiring privacy and data-protection scrutiny. IT and cyber security leads — technical controls, system use and security evidence. Digital or AI strategy leads — coordinating implementation, evidence and review.
No single role completes the whole assessment alone: each domain is owned by one role holder and read by a second, and specialist questions are referred out. AISG supports the professional judgement of the DSL, the DPO, governors and IT, security and legal advisers; it does not substitute for any of them, or for statutory guidance.
Beyond education
Many of the governance questions AI raises are not unique to schools: public-service organisations also have to consider accountability, privacy, security, human oversight, procurement, evidence and risk. The current AISG methodology is designed for education. Its underlying governance architecture may provide a basis for future sector-specific versions, subject to their own source mapping, controls, evidence requirements, critical-control analysis, specialist review and testing. No such version exists today, and nothing on this site should be read as a claim that AISG is suitable for other public services as it stands.
·
Contact
Wendell Gopaul Headteacher · Founder, AISG
AISG is developing through practitioner use, specialist feedback and methodological refinement. Enquiries are welcome from schools, trusts and organisations interested in exploring the methodology or contributing to its development.
AISG cannot advise on live safeguarding cases. If you have a concern about a child, follow your setting’s safeguarding procedure. Do not send case details to this address.
There is no contact form, because a form would transmit data and break the position that nothing leaves your device.
School pilot enquiries
For schools or trusts interested in exploring AISG or joining the pilot: include your school name, your role, and the term you would complete a return in.
Methodology feedback
For safeguarding, governance, privacy, legal, cyber security or research professionals interested in reviewing aspects of the methodology — the flagged-controls list on the methodology page is a standing invitation.
Professional collaboration
For education organisations, local authorities, public-service organisations or researchers interested in discussing the work.
Status
Application 0.5.8 · Methodology v0.5 · Control set v0.4 · Source register v0.4 · Scoring model v0.5. The methodology is reviewed annually, and between reviews where a source changes materially.
Pilot edition · Autumn term 2026
AISG is running as a pilot with schools for one term. The methodology is practitioner-developed and under active review. We are asking pilot schools to complete a return, tell us what does not work, and log the time it took.
To join, write to w.gopaul@st-annes.bham.sch.uk with your school name, your role, and the term you would complete the return in.
Prints as an A4 document. Choose “Save as PDF” for a copy, and set Headers and footers to off in the print dialog so no browser URL or date appears on the page.
AISGAI Safeguarding & Governance
In practice
AISG in practice: one school
How AISG was used in the setting it was written in, what it identified, and what this account cannot show.
Published
August 2026
Prepared against
KCSIE 2026, which comes into force 1 September 2026
Document ID
AISG-DOC-CASE-2026-08
Version
0.5.8
Standing note
This is one practitioner’s account of using AISG in their own setting. It is not evaluation evidence and does not show that AISG works generally.
About this case study
This case study documents the application of AISG within the school led by its creator. It provides a practical account of implementation, findings and learning from one school setting.
It is not presented as independent validation of the methodology; it is offered as evidence of practical application. Wider practitioner testing, specialist review and methodological development form the next stage of AISG.
The setting
[To be written: St Anne’s Catholic Primary School — phase, size, context, aggregated only. No pupil or staff detail.]
The starting point
[To be written: what was already in place before the review — policies, training, filtering arrangements — aggregated only.]
What prompted the review
[To be written: what triggered the first review and when.]
How AISG was used
[To be written: who completed it, over what period, how second review was handled.]
What the review identified
[To be written: the domains that came out weakest, aggregated. No control-level detail that identifies individuals.]
What actions followed
[To be written: the actions raised, owners assigned, and what has been closed.]
What worked
[To be written: which parts of the method earned their place.]
Limitations of this account
[To be written, and this section is what makes the page credible: what the review did not surface, where self-assessment bias was likely, and what a moderated review would have tested that this one did not.]
What changed in the methodology as a result
[To be written: the specific changes to controls, thresholds or wording that came out of using it.]
Next steps
[To be written: how and when the review will be revisited, and what a moderated (Tier 2) review would add.]
Plain-English explanations of the terms used across AISG. The same wording appears in the pop-up explanations throughout the site and in the printed key terms of the briefings, so a term never means different things in different places. These are explanations, not legal advice; the full treatment of each concept is in the and the .
AISGAI Safeguarding & Governance
Accessibility
Accessibility statement
How this resource supports access, and where its current limits are.
What we aim for
This site and the Evidence Review are built to meet WCAG 2.2 AA. Everything works with a keyboard alone, skip links are provided, and colour is never the only carrier of meaning — every rating also has a text label.
What is in place
All pages and the review itself are navigable by keyboard, with visible focus indicators.
The domain diagram carries a text description for screen readers, and is replaced by a plain list on small screens.
Documents print with structured headings so exported PDFs retain a reading order.
No motion, autoplay or time limits are used anywhere.
Known limitations
This is a pilot. Some generated documents have not yet been tested with assistive technology across all browsers. If something does not work for you, write to the contact address on the About page and we will treat it as a defect, not feedback.
Every partially evidenced or unevidenced control produces an action. Reports stay withheld until each action has a named owner and a target date.
Outputs
Reports
Generate audience-specific reports from the completed evidence review. Public-facing statements reflect the evidence recorded in this return.
How reporting works
Eleven documents come from one return. Each states that it comes from a Tier 1 self-assessment and has not been independently checked. In the staff, parent and pupil briefings, a sentence saying something is in place appears only where the control behind it is fully evidenced; otherwise the wording says the school is reviewing that area.
Set Headers and footers to off in the print dialog, so no browser URL or date appears on a board paper.To save as a PDF, choose Print and select “Save as PDF”. Reports print as A4 documents.
Methodology
Source register
What each core control rests on. A control is only called statutory where the primary source was checked; where it could not be checked, the row says so. Definitions of the classifications are on the page.
Methodology
Review methodology
A practitioner-developed, evidence-based self-review methodology, currently in pilot. This page sets out what AISG assesses, how every rating is produced, what it can and cannot claim, and how it is being tested. Nothing here changes how you complete the review; it is here so the operational screens can stay clean.
What AISG assesses
AISG structures a self-assessment of AI safeguarding and governance evidence across six domains: leadership and governance; safeguarding and pupil welfare; data protection and privacy; technology and procurement; teaching and staff practice; and filtering, monitoring, assurance and improvement. It produces a position the governing board can question, an action plan with named owners and dates, and stakeholder reports written in evidence-dependent language.
Methodology v0.5 · Control set v0.4 (58 core controls, 12 critical) · Source register v0.4 · Scoring model v0.5 · Jurisdiction: England · Status: pilot · Guidance basis: KCSIE 2026 and other relevant authoritative guidance in force at the review date · Review cycle: annual, following each new KCSIE edition, and after material regulatory or guidance change · Methodology current as of 18 August 2026 · Source register last verified 16 August 2026.
What AISG does not claim
AISG is not a certification, a legal opinion, a technical security audit, a statutory inspection or a validated assessment instrument. A rating summarises the evidence available against the controls in this methodology at the time of review. It does not establish statutory compliance, and it does not establish that safeguarding arrangements are effective in every circumstance. Self-assessment is not independent assurance.
Methodology provenance
AISG is a practitioner-developed methodology originating from school leadership practice. Its controls draw separately on authoritative sources, research evidence and explicitly identified practitioner methodology. The founder’s professional experience explains the origin and context of the framework; it is not presented as evidence that the methodology has been independently validated. ·
Designed to evolve with AI
AISG reviews the controls around AI use, not individual AI products. AI systems, their capabilities and the guidance surrounding their use will continue to change. Because the methodology is built around institutional controls rather than particular tools, it can evolve without the review model being replaced each time a new tool appears.
The review considers whether appropriate arrangements exist around safeguarding, governance, accountability, privacy, security, human oversight, evidence and monitoring. The sources and individual controls supporting those areas are then reviewed as legislation, guidance, technology and institutional practice change. AISG is not attempting to predict every future AI tool; it provides a structured governance layer through which changing technologies, guidance and institutional practices can be reviewed.
What may change, and what remains more stable
Likely to change
AI systems and capabilities
School use cases
Statutory and regulatory guidance
Data protection guidance
Safeguarding guidance
Technical standards
Procurement expectations
Available research evidence
More durable
Safeguarding responsibility
Institutional accountability
Human oversight
Privacy and data governance
Security
Risk ownership
Evidence of implementation
Monitoring, incident response and escalation
Review
The controls are written against the right-hand column. Changes in the left-hand column are absorbed by reviewing the affected sources, controls and evidence, not by rebuilding the methodology.
A change in an authoritative source or in the setting’s circumstances re-enters the chain at the point it affects. A change does not necessarily require the whole methodology to be rebuilt; it may require the affected sources, controls or evidence to be reviewed. The same chain gives traceability in the other direction: from any finding, back through the control and its required evidence, to the source or the stated methodological rationale — so a reviewer can always answer “why am I being asked this?” from the source register.
What can trigger a review?
Five trigger categories. These are AISG methodology concepts, not statutory categories.
Trigger
What happened
Possible outcome
Source change
Legislation, statutory guidance, regulatory guidance or another authoritative source changes.
Affected controls are identified and reviewed.
Technology change
A new AI capability introduces a materially different safeguarding, privacy, security or governance risk.
Relevant controls and evidence requirements are reconsidered.
Institutional change
The school introduces a new AI system or materially changes how an existing one is used.
The relevant parts of the school’s assessment are reviewed.
Evidence change
Evidence supporting a control becomes outdated, unavailable or no longer demonstrates current operation.
The control is marked “evidence review due”.
Incident change
An incident, near miss or emerging risk suggests an existing control may not be operating effectively.
Relevant controls are reassessed and actions recorded.
The three methodological layers
Every AISG finding rests on three separate judgements, and the methodology keeps them separate:
1 · Source authority
Why the issue matters: legislation, statutory guidance, regulatory or government guidance, technical standards, recognised professional guidance, or research evidence. Recorded per control in the source register.
2 · AISG control
How AISG has translated that requirement, risk or evidence into something a school can assess. The wording of an AISG control is not the wording of the underlying legislation or guidance unless the register quotes it.
3 · AISG assessment rule
How AISG decides whether evidence is sufficient, whether a control is demonstrated, whether it is critical, how it contributes to a rating, and when specialist review is required. These rules are AISG methodological choices, published on this page.
Source authority establishes why a control matters. Evidence strength establishes how well its operation is demonstrated. The AISG assessment model determines how those findings are summarised. These are separate judgements.
How controls are derived, and how sources are classified
Each control carries a classification describing the status of the AISG control — which is not the same thing as the status of the source that informs it. Where a practitioner control is informed by a statutory or guidance source, the register records both (control status P, source basis S or G), so AISG wording is never implied to appear in legislation.
Statutory
Statutory or authoritative requirement. Grounded in legislation, statutory guidance, regulation or another applicable authoritative duty. KCSIE is statutory guidance schools must have regard to, so a duty KCSIE states with “should” is treated as statutory.
Official guidance
Authoritative guidance or expected practice. Governmental, regulatory or recognised professional guidance that informs expected practice but is not represented by AISG as a direct statutory requirement.
Practitioner control
Practitioner-developed control. Developed by AISG to operationalise risk management, evidence gathering or good governance. No statute or guidance requires it, though the register records any source that informs it.
The three classifications do not have equivalent legal status. Legal status is never determined from words such as “must”, “should” or “may” alone; context and the underlying authority matter. Where authority or legal status is uncertain, the classification is downgraded rather than elevated, pending specialist verification. Uncertain classifications are flagged for legal or safeguarding review in the register and at the foot of this page; AISG does not resolve uncertain legal interpretations itself. The S/G/P abbreviations remain in exports and the source register.
Tier
Source hierarchy
A
Primary authority: legislation, statutory guidance, DfE, ICO, NCSC, UK government, other statutory or regulatory sources.
B
Official implementation guidance: DfE standards, UKCIS, government-supported implementation resources, official sector guidance.
C
Standards and governance frameworks such as ISO/IEC 42001 and NIST AI RMF, only where relevant.
D
Peer-reviewed research. Informs the methodology. Never creates a legal obligation.
E
Practitioner methodology: controls this framework introduces because they support good governance.
The authoritative-source register records, per source: title, authority, date, the exact wording relied on, whether the control goes beyond the source, and verification status. Wikipedia is used only to locate original sources; it is never cited as the basis for a control. Recording a source URL and effective date for every register entry is on the roadmap.
Evidence model
The model separates two things: evidence maturity (how well the evidence demonstrates the control) and assurance level (who has checked the judgement). The governing principle: a document existing does not, by itself, demonstrate that a control operates in practice. A written policy shows a control has been designed; the maturity scale below lets the review consider not only whether arrangements are documented, but whether there is evidence they are being implemented.
Evidence maturity (E0–E3)
E0 — Assertion only
A claim has been made, but no supporting evidence has been identified.
E1 — Evidence identified
Relevant documentary or other evidence has been identified by title and date.
E2 — Evidence reviewed
Someone in the setting has examined the evidence, and it reasonably supports the existence and design of the control.
E3 — Operational evidence
Evidence that the control operates in practice: minutes, training records, logs, a completed review, an inventory, a decision record, testing or monitoring records, incident handling, a trend review, or a sample assurance record.
Assurance level
Recorded separately from maturity: self-assessed (Tier 1), internally moderated (second review within the setting), independently reviewed (Tier 2 and 3). Independent review adds assurance over the judgement; it is not automatically stronger than operational evidence, because the two measure different things.
Compatibility note. Earlier versions expressed independent review as a fifth maturity level, E4. E4 is retained in the saved-return format so older returns open unchanged, but it is read as an assurance level, not a maturity level, and cannot be selected in a self-assessment. Removing E4 from the maturity scale entirely is on the methodology roadmap.
What “fully evidenced” means
A control is fully evidenced when the evidence supplied meets or exceeds the minimum evidence threshold specified for that control. Every control has a defined minimum:
Control type
Minimum for fully evidenced
Critical controls, and controls about operational practice (16 controls, listed in the source register)
E3 — operational evidence, plus an evidence title and date
All other controls
E1 — evidence identified by title and date
The threshold reflects the nature of the control; not every control requires operational evidence. The application shows each control’s minimum beside the evidence-strength field, and no report is produced while a rating sits below its threshold. Whether the E1 minimum for standard controls should rise to E2 is under review in the pilot.
How ratings are calculated
The calculation is published so it can be reproduced by hand from a completed return.
Fully evidenced percentage = controls rated fully evidenced ÷ applicable assessed controls × 100, per domain, where each state is treated as follows:
State
Treatment in the calculation
Fully evidenced
In the denominator; counts towards the percentage. Must meet the control’s minimum evidence threshold before the return is treated as complete.
Partially evidenced
In the denominator; does not count towards the percentage. Generates an action.
Not evidenced
In the denominator; does not count towards the percentage. Generates a high-priority action, or a critical one where the control is critical.
Referred for specialist review
Excluded from the denominator until the outcome is recorded; stays visible in every output.
Not applicable
Excluded from the denominator. Never counts as green. Requires a written reason; visible in every report; challengeable at moderation.
Not yet assessed
Excluded from the denominator, and blocks the domain rating under the completion rule below. Leaving difficult controls unanswered can never improve a score, because no rating is presented until they are answered.
The percentage maps to a rating, with the critical-control override applied first:
Rating
Rule
●Green
80% or more of assessed controls fully evidenced, and no critical control not evidenced, and no more than one critical control partially evidenced.
▲Amber
50–79% fully evidenced, or two or more critical controls partially evidenced.
■Red
Under 50% fully evidenced, or any critical control not evidenced.
○Grey
Not enough assessed to rate, or referred out.
○Incomplete
One or more applicable controls in the domain not yet assessed. No rating is presented.
The RAG thresholds are practitioner-defined pilot thresholds. They are not DfE, Ofsted or statutory thresholds. They will be reviewed once pilot data and outside practitioner comment are in. They are not scientifically validated. Which controls are marked critical is this methodology’s judgement; no authoritative source designates any control critical. High scores elsewhere never compensate for a critical safeguarding failure.
Minimum completion rule
A domain rating is calculated and presented only once every applicable control in the domain has been assessed — rated, referred with a complete referral, or marked not applicable with a reason. Until then the domain shows “Assessment incomplete” with a count of controls assessed, and the whole-setting position reads as incomplete. Two passing controls out of ten can never produce a green. Critical controls already rated not evidenced are surfaced immediately, without waiting for completion. This rule was added in methodology v0.5; the thresholds themselves are unchanged.
What a rating means — and what it does not
An AISG rating summarises the evidence available against the controls included in the methodology at the time of review. It does not establish statutory compliance or guarantee safeguarding effectiveness.
A green rating means the applicable controls met the defined AISG evidence threshold and no critical-control rule prevented a green result. It does not mean zero AI risk, guaranteed compliance, guaranteed safeguarding effectiveness or permanent approval, and it does not make future reassessment unnecessary. This wording appears beside every overall rating the application produces.
Critical controls
Definition. A critical control is a control whose absence or failure could leave a material safeguarding, governance, privacy, security or accountability risk without an effective preventive, detective or responsive control, and where stronger performance elsewhere would not reasonably compensate for that absence.
If a critical control is not evidenced, the domain is rated red whatever else is in place. For every critical control the source register records: the critical status; the rationale for the designation, stating the specific foreseeable risk left unmanaged; its provenance (practitioner-defined unless the register states otherwise — no authoritative source designates any control critical); the required evidence threshold (E3); the consequence of failure; and any specialist-review flags. A reviewer should be able to answer, from the register alone, why one control is critical and another is not.
12 of the 58 core controls are critical. Controls are not added to this list to make the methodology appear stronger; v0.4 removed one designation (D1.1) on exactly that reasoning.
From review to action
The purpose of the review is not the score. It is to identify what can be evidenced, where evidence is incomplete, where controls need attention and where specialist judgement is needed, and to turn each finding into an action with an owner and a date rather than ending at a RAG rating.
Finding→Action→Owner→Target date→Evidence of completion→Review
This chain is implemented: every partially evidenced or unevidenced control generates an action with an owner, target date, status and evidence-of-completion field, and reports are withheld until every action is owned and dated.
Evidence gaps
An evidence gap is not automatically a control failure. The methodology distinguishes: control absent (rated not evidenced); control present but insufficiently evidenced (rated partially evidenced, or held below its minimum threshold); evidence unavailable or pending (the rating is withheld and the domain stays incomplete); not applicable (with a written reason); and specialist judgement required (referred). Reports declare gaps explicitly: an unassessed domain is reported as an evidence gap, never filled with prose, and a lack of evidence is never converted into a conclusion.
Not applicable
N/A is a legitimate outcome where a control genuinely does not apply. It requires a written reason, records who made the decision and when, is excluded from the denominator, never counts as green, stays visible in every report, and can be challenged at moderation. For a critical control, the reason must also explain how the underlying risk is otherwise managed, and the second reviewer checks the decision. N/A cannot be used to remove a difficult control from scoring: exclusion from the denominator means it never improves a percentage.
Specialist referrals
Some controls need a specialist to settle: safeguarding, DPO and privacy, legal, IT and cyber security, filtering and monitoring configuration, procurement, HR or governance questions. Where a control requires specialist interpretation it is marked referred, rather than AISG making a confident determination outside its competence. A referral preserves the original control question and records the specialist, the reason, an owner and a target date; the control stays visible until the outcome is recorded. Referring a control does not close it.
Review and moderation
Reviewer 1 — the domain role holder — completes the assessment: ratings, evidence review, evidence-level judgements, N/A decisions and referrals.
Reviewer 2 — the named second reviewer — checks, as a minimum: every critical control, every N/A decision, every specialist referral, every red or high-priority finding, and any evidence judgement they doubt, plus a sample of the remaining controls. At Tier 2, all remaining controls.
Disagreement. Both initial judgements are recorded: the second reviewer’s rating is stored alongside the primary rating, never over it. The reviewers reassess against the published evidence criteria on this page, record the agreed final judgement as the primary rating with the change noted in the second-reviewer comment, and both originals remain in the export so the audit trail is preserved. Disagreement rates are captured in the pilot feedback export for future reliability analysis.
Second review reduces optimism and self-report bias; it is not independent assurance. Tier 2 (moderated) and Tier 3 (independent) reviews add assurance over the assessment process. They are not certification or accreditation, and they do not guarantee safeguarding effectiveness or legal compliance.
Evidence currency
Evidence ages. Every evidence entry carries a date, and every return a completion date. The defined review period is the annual governance cycle: operational evidence more than 12 months old is flagged “evidence review due” on the overview, and a material change to a tool, its terms, functionality or data handling triggers proportionate re-review sooner. Historical operational evidence is never assumed to show a control continues to operate indefinitely. Per-control review periods shorter than the annual cycle are deliberately not invented; if pilot data shows a control type needs one, it will be defined by control type, not arbitrarily.
Methodology versions
Every saved return and every export is stamped with the methodology version, control-set version, source-register version, scoring-model version and the review date. Historical returns continue to show the versions under which they were completed; they are never silently re-scored against a later methodology. Opening an older return in a newer edition preserves its ratings and shows a notice. Applying a new methodology retrospectively is a reassessment, and is shown as one.
Current: Methodology v0.5 · Control set v0.4 · Source register v0.4 · Scoring model v0.5.
Provenance: what each element rests on
Where an element of this methodology is an AISG choice rather than an external requirement, this table says so.
Element
Source / authority
What the source supports
What AISG adds
Safeguarding controls
KCSIE 2026 and related statutory guidance
The underlying safeguarding duties
AI-specific assessable controls and evidence questions
Privacy and data protection
UK GDPR, Data Protection Act 2018, ICO guidance
The data-protection framework and DPIA expectations
AI-specific evidence questions; the review does not perform a DPIA
Filtering and monitoring
KCSIE 2026 (para 171) and DfE standards
The annual review expectation and the standards
AI-specific coverage checks
Technology and procurement
DfE product safety standards, NCSC guidance
Expected practice for adoption and security
Adoption, decommissioning and vendor-change controls
Leadership and accountability
Research and governance guidance (see references)
The need for named accountability and oversight
The leadership domain and its controls
Human oversight and AI literacy
DfE guidance; research (see references)
Professional accountability for AI-assisted work
Assessable oversight and literacy controls
Evidence maturity (E0–E3)
General assurance and governance principles
Distinguishing design from operation
The specific scale and per-control thresholds — an AISG choice
RAG thresholds (80% / 50%)
None — practitioner methodology
—
The thresholds are AISG judgements under pilot review; no research validates them
Critical-control override
Risk-management principle: material failures are not averaged away
The principle
The override rule and which controls are designated — AISG judgements
Moderation and tiers
Moderation practice in education assessment
The value of second review
The specific protocol and tier definitions
No research paper validates AISG scoring thresholds, and none is claimed to.
Research and evidence base
Two source layers are kept apart and never merged into one bibliography. Authoritative sources — legislation, statutory guidance, regulator and government guidance — establish applicable duties, expectations and guidance; the source register maps them control by control. Research literature informs the broader methodology — governance, leadership, ethical implementation, human oversight, institutional readiness. Research references do not by themselves establish statutory requirements, do not create duties, and do not validate the scoring model. Each reference states which principle it informs.
Verified against the publisher record:
Bond, M., Khosravi, H., De Laat, M., Bergdahl, N., Negrea, V., Oxley, E., Pham, P., Chong, S. W., & Siemens, G. (2024). A meta systematic review of artificial intelligence in higher education: a call for increased ethics, collaboration, and rigour. International Journal of Educational Technology in Higher Education, 21, art. 4. DOI 10.1186/s41239-023-00436-z. Informs: the need for explicit ethical consideration, governance and transparency in institutional AI adoption. Higher-education scope; principles only.
Wieczorek, M., Hosseini, M., & Gordijn, B. (2025). Unpacking the ethics of using AI in primary and secondary education: a systematic literature review. AI and Ethics, 5(5), 4693–4711. DOI 10.1007/s43681-025-00770-0. Informs: school-specific ethical risks — privacy, fairness, oversight — underlying the safeguarding, privacy and human-oversight domains.
DeMatthews, D., Reyes, P., Hart, T. D., & James, L., III (2026). Leadership for artificial intelligence use in schools: a six-domain framework. Educational Management Administration & Leadership, advance online. DOI 10.1177/17411432261418940. Informs: the six-domain idea only. US K–12 context.
Borgonovi, F., Bastagli, F., Ochojska, M., & Piumatti, G. (2025). AI adoption in the education system. OECD Artificial Intelligence Papers No. 52. DOI 10.1787/69bd0a4a-en. Informs: policy direction on institutional AI adoption. Scoped to Italy.
Identified, not yet verified against the publisher record, and therefore not relied on by any control or rule: Bellibaş & Karaferye (2026), Improving Schools, on leadership accountability; Karaköse et al. (2026), Frontiers in Research Metrics and Analytics, on governance and human-centred oversight; Ali (2026), Management in Education, on leadership responsibility and ethical judgement; Bagdonaite & Dagiene (2025), Informatics in Education, on the primary-education context; Zulfiqar & Iqbal (2026), Education Innovations, on responsible implementation in primary settings; Alyoussef et al. (2026), Frontiers in Artificial Intelligence, on teacher AI-literacy competencies. Each will be cited only after verification, with the principle it informs stated alongside.
Limitations
AISG is a practitioner-developed methodology currently in pilot. It has not been established as a validated psychometric or regulatory instrument.
Ratings summarise evidence against the controls included in this methodology. They do not establish legal compliance, and they do not establish that safeguarding arrangements are effective in every circumstance.
Self-review introduces reporting and confirmation bias. Second review reduces it; it does not remove it.
Evidence quality depends partly on the evidence the institution supplies. A document existing does not show a control operates.
Not every risk can be found by reviewing documents.
The RAG thresholds are judgement-based pilot thresholds, not statistically derived, and inter-rater reliability has not yet been measured.
AI technology and the guidance around it continue to change; source mappings are correct as at the register date, not indefinitely.
Legal, data-protection and cyber-security questions need specialist assurance. The methodology refers them out rather than settling them.
A point-in-time review does not guarantee ongoing effectiveness.
Development and validation roadmap
What has to happen, in order, before AISG could reasonably claim more than it claims today. None of it has happened yet, and the methodology says so.
Stage 1 — Practitioner pilot Current stage
Use in real school settings. Collect completion time, usability feedback, control ambiguity, evidence availability, referral and N/A frequency, and rating distributions — via the pilot feedback and time log, exported separately from the school’s return.
Stage 2 — Specialist review
Structured review by safeguarding, DPO and privacy, cyber-security, governance and education-leadership specialists, and legal specialists where required. The flagged-controls list at the foot of this page is the standing input.
Stage 3 — Inter-rater reliability
More than one reviewer independently assesses the same evidence; agreement is measured on evidence levels, control status, N/A decisions and critical findings (percentage agreement, with kappa where appropriate). The data model already preserves both reviewers’ judgements. No reliability claim is made until it is measured.
Stage 4 — Threshold review
Examine whether the green, amber and red thresholds and the critical overrides produce useful, defensible distinctions. Thresholds will not be changed to obtain a desirable share of green schools.
Stage 5 — External methodological review
Independent scrutiny of the structure, evidence model, scoring, source mapping, critical controls, limitations and reporting.
Also on the roadmap: removing E4 from the maturity scale with assurance recorded separately end to end; recording a source URL and effective date for every register entry; and reviewing whether the E1 minimum for standard controls should rise to E2. Register entries will gain machine-readable source-currency fields (source version, publication and effective dates, last verified, URL, authority type, controls affected), so that when an authoritative source changes, the affected controls can be identified for review — identified, never automatically rewritten: a changed webpage does not reclassify a control without human review. Governance areas expected to matter more as AI use becomes operational — procurement and supplier assurance, AI-supported decision-making, explainability and decision records, equality impacts, data provenance, model and system change, incident reporting, transparency, staff AI literacy, monitoring, and system decommissioning — will each be source-mapped and specialist-reviewed before becoming a control, and none will be made critical by default. The long-term architecture is a sector-neutral governance core with sector control packs, education first; no non-education pack exists, and none is claimed.
Methodology changelog — v0.5
Separated and published the three methodological layers: source authority, AISG control, AISG assessment rule.
Clarified evidence maturity as E0–E3 and separated assurance level from maturity; E4 retained for compatibility and reframed as assurance.
Defined “fully evidenced” against a published per-control minimum evidence threshold (unchanged in effect; now explicit).
Published the exact scoring calculation and denominator rules.
Added the minimum completion rule: no domain rating until every applicable control is assessed. The one change to what is presented; thresholds unchanged.
Formalised the critical-control definition and its required rationale fields.
Strengthened N/A: who and when now recorded; critical N/A must address how the risk is otherwise managed.
Formalised the moderation protocol and disagreement handling.
Added the provenance matrix, expanded limitations and the validation roadmap.
Verified research references against publisher records and connected each to the principle it informs; unverified candidates held pending.
Added the living-methodology explanation: what changes and what remains stable, the five review-trigger categories, and the designed-to-evolve positioning.
These changes are methodology refinement, not validation. Controls, their wording, classifications, RAG thresholds and critical designations are unchanged from v0.4. Returns saved under earlier versions open unchanged and are never re-scored. Earlier versions are in the .
Controls flagged for human review
Generated from the source register. These are the controls a specialist should look at before the methodology moves past pilot.
Methodology
Moderation & tiers
A self-assessment cannot give a board independent assurance. The people filling it in are often the people who chose the tools. The three tiers let a school bring in more independence when it wants it.
Tier 1 — Self-Assessment
The setting completes it. Evidence is declared or internally reviewed. No independent verification. This edition.
Tier 2 — Moderated Evidence Review
An independent practitioner reviews selected evidence and challenges the ratings, may amend findings, and issues a moderation note. Not legal certification.
Tier 3 — Independent Health Check
An independent practitioner examines evidence, conducts structured interviews and makes the assessment, with action planning and a governor briefing. Still not legal certification, statutory inspection, a technical cyber audit or a guarantee of safety.
Moderation protocol
Reviewer 1 (the domain role holder) completes the assessment: ratings, evidence review, evidence-level judgements, N/A decisions and referrals. Reviewer 2 (the named second reviewer) checks, as a minimum: every critical control, every N/A decision, every specialist referral, every red or high-priority finding, and any evidence judgement in doubt — plus a sample of the rest (all of them at Tier 2).
Disagreement. Both initial judgements are recorded — the second reviewer’s rating sits alongside the primary rating, never over it. The reviewers reassess against the published evidence criteria, record the agreed final judgement with the change noted in the second-reviewer comment, and both originals stay in the export so the audit trail is preserved.
Export for Tier 2 moderation
The export holds your ratings, evidence references, referrals and action plan. Nothing is sent automatically. You decide what goes, to whom, and how.
Send the exported file by a secure route to a named person. This file describes where your governance is weak, so it should not go over consumer messaging.
Before exporting. This file describes where your governance is weak. Send it securely to a named person, and check first that no child-level or staff-allegation information has crept in anywhere.
Administration
Setting details
Each domain is owned by one role holder and read by a second. That makes it harder for one optimistic person to rate the school alone. Nothing you type is sent anywhere; the return stays on this device until you save or export it.
1
Setting details
Who this return belongs to. These details appear on every report and export.
2
Review team and domain ownership
The second reviewer is there to catch what the first person missed or rated too kindly. It is not independent assurance.
3
Local controls (optional · additive only)
You can add controls for your trust, diocese or setting. The 58 core controls, their wording and the thresholds are locked, so one school’s return can be compared with another’s. Anything you add is marked as local wherever it appears.
4
Before you start
The review looks at systems and procedures, never at cases.
Do not enter names, safeguarding case details, case chronology, identifiable pupil information or identifiable staff allegation information anywhere in this tool. Evidence should be about systems and procedures, or aggregated and anonymised. This review does not look at cases.
If a live safeguarding concern arises during the review
Stop discussion of the individual case within the review.
Refer immediately through the setting’s safeguarding procedure.
Normally that means the DSL.
If the concern relates to the DSL or a member of staff, follow the applicable allegations or escalation procedure instead.
Do not record child-level substance in this application.
Record only that an appropriate safeguarding escalation occurred, where necessary.
The Evidence Review does not investigate safeguarding cases.
Self-assessment ·
Administration
Pilot feedback and time log
This screen is optional. It exports on its own, separately from the school’s return, and is used to improve the method. It is never used to score the school and is not part of the governance record.
Time by domain (minutes)
Feedback by domain and by role
Administration
Save & open a return
Your work is kept in this browser on this device as you go. Saving writes a portable file you can keep or hand on; opening reads one back. Nothing is uploaded and nothing is transmitted.
Save this return
Your work is kept in this browser as you go. Saving a file gives you a copy you can keep, move to another device, or hand to a colleague.
No changes yet
Open a saved return
Drag a saved return here
or
JSON returns saved by this tool (.json) · up to 8 MB · opening replaces whatever is currently on screen.
On a shared or public computer, clear the local draft when you finish. The working copy stays in this browser until you clear it or start again.
Administration
Data & versions
What this tool stores, what it sends, and how the methodology has changed.
What this tool stores and sends
Stored: a working copy of your return, in this browser’s local storage, on this device only. It is there so you do not lose work if the tab closes. You can clear it at any time.
Sent anywhere: nothing. No analytics, no cloud storage, no external fonts, no external transmission of any kind.
Exported: only when you press Save, Export for moderation, or Export pilot feedback. Each downloads a file to your device.
On a shared or public computer, clear the local draft when you finish.
Version history
Version
Change
v0.1
Initial practitioner prototype: six domains, 51 controls, evidence-led RAG with critical-control override, ten stakeholder reports, local additive controls.
v0.2
Renamed to Evidence Review, Self-Assessment Edition. Each core control now carries an S, G or P basis and a source register entry. Wording that could be read as assurance was changed. Evidence is entered as title, date, owner and note, and marked declared or reviewed internally. Countersignature became second review. Safeguarding warning and live-concern steps expanded. Referrals now record specialist, reason, owner and date. Actions carry a status, and reports are held back until every action has an owner and date. Parent, pupil and staff wording is now conditional on the evidence. Pilot feedback and time log added, exported separately. Every export is stamped with the versions used. D5.9 on intellectual property added.
v0.3
Statutory and guidance precision: six controls reclassified S to G where the source did not support S (D1.2, D1.5, D1.7, D3.4, D4.7, D5.6). DPIA, DPO, criminal-law and intellectual-property wording corrected so no obligation reads stronger than its source. Evidence-quality model E0 to E4. Five controls added: incident response, shadow AI discovery, vendor change monitoring, AI-assisted safeguarding systems (critical), pupil voice. Decommissioning, vendor review, human oversight, AI literacy and continuous improvement expanded. Source register extended with source tier, source wording, beyond-source flag and human-review flags. Inter-rater capture, limitations and research note added. 57 controls, 13 critical.
v0.4
Source-precision release. Seven controls reclassified S to G under the deletion test (D1.6, D2.4, D2.5, D2.6, D2.7, D4.3, D5.8): where removing the AI-specific words leaves a general statutory duty and the AI implementation is ours, the control is G. D5.5 raised P to G on the DfE’s stated position on professional accountability. D4.9 and D4.10 now G-supported. D1.1 no longer critical: a P-basis control should not cap a domain on its own. D6.3 split: the para-171 device check stays S; the AI-accessibility consideration is D6.3a, P. D4.2 updated to the renamed DfE product safety standards (January 2026). Seven source mappings verified against primary text on gov.uk. Not applicable added as a fifth state, reason required, excluded from the denominator, never green, visible in reports, challengeable at moderation. Critical controls now need operational evidence (E3) for a full rating, without exception. 58 controls, 12 critical, S 16, G 29, P 13.
v0.5
Design release. New application shell, navigation, overview, document centre and print design. No methodology changes: controls, classifications, thresholds, critical-control logic, evidence rules, report logic and the saved-return format are unchanged from v0.4. Returns saved under v0.4 open in this edition.
v0.5.1
Stakeholder gateway. A front page with plain-English briefings for staff, senior leaders and governors now sits in front of the Evidence Review, with “Briefings & home” available from the review navigation. Presentation layer only: no methodology, control, source, report or data-format changes.
v0.5.2
Methodology transparency release: methodology v0.5, scoring model v0.5. The three methodological layers (source authority, AISG control, AISG assessment rule) separated and published. Evidence maturity clarified as E0–E3 with assurance level recorded separately; E4 retained for saved-return compatibility and reframed as an assurance level. “Fully evidenced” defined against a published per-control minimum evidence threshold (E3 for critical and operational controls, E1 otherwise — unchanged in effect, now explicit). Exact scoring calculation and denominator rules published. Minimum completion rule added: a domain rating is presented only once every applicable control has been assessed; until then the domain reads “Assessment incomplete”. This is a flagged change to what is presented, not to the thresholds. N/A decisions now record who and when, with stronger justification wording for critical controls. Moderation protocol and disagreement handling documented. Evidence older than 12 months flagged for re-review. Provenance matrix, expanded limitations and validation roadmap added; research references verified against publisher records and connected to the principles they inform. Controls, wording, classifications, RAG thresholds and critical designations are unchanged from v0.4; returns saved under earlier versions open unchanged and are never re-scored.
v0.5.3
Living methodology explainer. Methodology page gains: Designed to evolve with AI; what may change and what remains more stable; the living methodology model with source-to-control-to-evidence traceability; the five review-trigger categories (AISG concepts, not statutory); what a rating means and does not mean; from review to action. Homepage gains a “governance that can keep pace with change” section; About gains “Who AISG is for” and a deliberately restrained “Beyond education” statement. Roadmap extended with source-currency fields and future governance areas. Presentation and documentation only: no methodology, control, scoring, source or data-format changes; existing assessments unaffected.
v0.5.4
Trust and provenance release. Homepage gains a restrained “Developed from practice” section linking to About and the case study. About restructured: why AISG exists → built from school leadership practice → the leadership-and-governance pairing (responsive on mobile) → founder → enquiry routes for school pilots, methodology feedback and professional collaboration. Case study gains an “About this case study” disclosure separating practical application from independent validation, plus starting-point and next-steps sections. Methodology page gains a provenance statement. Presentation only: no methodology, control, scoring, source or data-format changes.
v0.5.5
Homepage editorial release. Hero rewritten in evidence language (“what can be evidenced, where gaps remain and what may need attention”) with a practical facts line and direct start/open actions; briefings and outputs explicitly separated as before and after the review; the eleven outputs listed with the three most significant first and the remainder behind a keyboard-accessible expander (all stay in the page); evidence-aware reporting presented as a labelled methodology note; the repeated six-domain list removed from the closing call to action; pilot copy clarified (“pilot use is not validation”) and version numbers linked to the methodology, source register and changelog. Copy and layout only: no methodology, control, scoring, source or data-format changes.
v0.5.6
Briefing design release. The three role briefings gain explanatory graphics — a check–protect–verify–escalate flow and a “stop and ask” note for staff, and a board assurance cycle plus an oversight-versus-operations panel for governors (leaders keep the governance-cycle diagram) — and accessible key-term definitions that work by mouse, keyboard focus and touch, dismiss with Escape, and are repeated in a printed “Key terms” glossary in each briefing, numbered from the marked terms. Printed and exported briefings carry every definition and graphic without interaction, in colour or grayscale, with no external assets. Briefing content otherwise unchanged; no methodology, control, scoring, source or data-format changes.
v0.5.7
Review-panel refinements. Hero explains what a control is in plain words; the staff briefing tells the DSL a safeguarding extract exists for them; the governor briefing notes that its five questions become school-specific in the generated board briefing; the source register table is keyboard-focusable with a screen-reader label for its sideways scroll; mobile hardening (no page-level horizontal panning, internally scrolling tables, minimum 24px tap targets). Copy and accessibility only: no methodology, control, scoring, source or data-format changes.
v0.5.8
Site-wide key-term system. A single terminology register now drives every explanation: pop-up definitions work by hover, keyboard focus and touch, stay open while the pointer moves onto them, pin on click, close on Escape, and carry a “why it matters” line and a link to the new Glossary page (linked from the footer, printed from the same register). Each review domain header offers the six methodology terms used on that screen — evidence strength, operational evidence, critical control, specialist referral, not applicable, evidence gap — so nobody has to leave a control to understand it. Printed briefings keep their numbered key-terms glossaries; pop-up UI never prints. Explanatory only: no methodology, control, scoring, source or data-format changes.