Safeguarding statement

What AILitKit is, and what it is not

AILitKit is an adult-only professional planning tool for teachers and school leaders. It generates AI literacy guides from your curriculum context (subject, key stage, topic, scheme of work). It is not a student-facing service, not a safeguarding case-management system, not intended to receive pupil personal data, and not a substitute for your school's designated safeguarding lead (DSL). Every guide is for teacher review before classroom use.

Safeguarding contact and how to raise a concern

If something the platform produced concerns you, if you believe a request was wrongly refused, or if you want to flag a pattern of activity to us, contact us using any of the routes below. We aim to acknowledge within 1 working day and respond within 5 working days.

• Safeguarding inbox: safeguarding@ailitkit.com for any concern about a generated guide, a classifier decision, or content that has surfaced through the product.
• Appeals and human review: the same inbox, with the date, subject, key stage, and topic. A human reviews the decision, takes your point of view into account, and replies. This route is also documented in our Privacy Policy under “Automated decisions and how to appeal”.
• General questions: hello@ailitkit.com.

A named AILitKit safeguarding liaison reviews every message to the safeguarding inbox. We are not a regulated safeguarding service and do not act as a DSL for your school. Where a message describes an active safeguarding concern about a pupil, we will direct you to your own DSL, the NSPCC helpline (0808 800 5000), or the police as appropriate.

How the safeguarding gate works

Before new generation, automated safeguarding checks read the subject, stage, year group and topic, plus up to the first 1,000 characters each of the teacher description and extracted upload text. Llama Guard gives a first verdict; a Gemini reviewer runs if that verdict is unsafe, unavailable or unreadable. An additional direct OpenAI moderation check runs when configured, under its separate API terms. If no usable safeguarding verdict is available, generation pauses with a retry message. A usable unsafe verdict is preserved if its reviewer fails; hard-block categories cannot be cleared by that reviewer.

Our OpenRouter account has zero data retention (ZDR) enabled, and requests also require ZDR inference endpoints. This applies to provider processing of prompts and outputs, including retries and fallback. It does not delete your saved AILitKit guides, account records or necessary service metadata. OpenRouter permits temporary in-memory prompt caching under its ZDR policy.

The sampled input check does not cover every character of long source documents. The drafting rules and automated content review reduce risk but cannot guarantee safe output; teachers must review the final guide.

What the three verdicts mean

• ALLOW. The sampled request passed the automated gate. The guide is generated normally and no input snippet is stored in the safeguarding log (allowed requests are not reviewed by humans).
• SENSITIVE. The topic engages a safeguarding-sensitive area within a legitimate curriculum frame, for example PSHE drugs education, KS5 Biology drug pharmacology, RSHE, FGM awareness, or Prevent. SENSITIVE is not a warning about the teacher. It records that the system has applied a safeguarding-aware framing (awareness over instruction, age-appropriate depth, DSL signposting) to the generated guide. The verdict, category, model, and the topic and up to 250 characters of free-text are stored in the safeguarding log; uploaded files themselves are not.
• BLOCK. The request was refused before any guide was generated. You see a message explaining why and how to appeal. The verdict, category, model, and the topic and up to 250 characters of free-text are stored.

Hard-block categories

A Llama Guard hard-block verdict for either of these categories cannot be cleared by the automated curriculum reviewer:

• Child sexual exploitation (MLCommons S4): including any request that would describe, depict, or instruct the generation of child sexual abuse material.
• Indiscriminate weapons (MLCommons S9): chemical, biological, radiological, nuclear, and explosive weapons of mass destruction.

These classifications can be wrong. The human appeal route remains available if legitimate educational content was misclassified. A content flag is not evidence of teacher misconduct.

What your school admin sees

If you joined AILitKit through a school or trust account, your school administrators (and trust administrators where applicable) can see safeguarding decisions on accounts in their organisation. The view is limited:

• Admins see SENSITIVE and BLOCK rows for staff in the organisation, not ALLOW rows. Routine activity is not visible.
• Each row shows the teacher's name, the verdict, the category, and the first 120 characters of the topic or free-text description.
• The log retains the topic and up to 250 description characters, plus the reason. The 120-character display is only a presentation limit, displayed truncated at render time. It is not a separate stored copy.
• Admins cannot read the contents of your guides or the contents of any uploads.

The purpose is to let the designated safeguarding lead step in if a colleague is repeatedly hitting the gate, in the same way a school filtering and monitoring solution surfaces patterns to the DSL under KCSIE Part 2. If you write a request that engages a sensitive curriculum topic, expect your school admin or DSL to be able to see that you did. AILitKit operations staff also have a service-role view across all organisations for support and abuse investigation; use of that view is logged to the administrative-actions audit, retained for 365 days.

Safeguarding controls built into every generated guide

When the curriculum-aware reviewer returns SENSITIVE, AILitKit injects a safeguarding addendum into the system prompt before the guide is generated. The addendum instructs the model to follow these rules; teachers must check that the resulting guide does so:

• Awareness, not instruction. The guide covers recognition, prevention, support, and where to get help. It must not include step-by-step methods, dosages, sourcing routes, or anything that could enable harm to self or others.
• Age-appropriate framing. Depth and language match what is statutory at the selected key stage; the guide does not pre-empt content from later key stages.
• DSL signposting. Any activity that may surface lived experience or invite disclosure includes a teacher coaching note about your school's DSL referral pathway.
• Teacher-led by default for the most sensitive elements, rather than open student-led activities, unless the framing is explicitly low-risk.
• Statutory reference. KCSIE 2026 is referenced in the governance footer for any Part 1 / Part 5 concern, including its provisions on AI-generated imagery, deepfakes, and AI-simulated interaction.
• Category-specific guidance. Each category (self-harm, RSHE, FGM, Prevent, drugs, knife crime, genocide and war crimes, religion and belief, science of reproduction, science of drug action) has a tailored rule. For example, eating disorders are awareness-only with no specific behaviours, weights, or strategies that could be imitated.

Retention of safeguarding records

Safeguarding-classifier decisions are retained for the life of the underlying account and removed by database cascade on account deletion. ALLOW rows hold no input snippet. SENSITIVE and BLOCK rows hold the topic and up to 250 characters of the free-text description and a boolean flag indicating whether an upload was involved. The original file and full extracted upload body are not stored in this log; the classifier explanation may refer to screened material.

Teacher-led delivery

Before saving a newly generated guide, automated checks review lesson fit, accuracy, activity logic, resources, pupil language and AI literacy. Up to four batches inspect the draft against the teacher brief; one correction cycle is allowed, followed by another review of all batches. Failed or unavailable review prevents the new guide from being saved. Saved review evidence and resolved findings may quote the brief or draft. These checks can miss errors and are not human approval or a dedicated final-output safety classifier. Existing guides are not retrospectively checked; eligible saved guides may be reused.

Pupil-facing language is targeted using the year group or key stage supplied by the teacher. This is a class-level assumption, not an individual reading-age assessment. Teachers must adapt wording and resources for their actual readers, including SEND and EAL needs, and check third-party tools, links and safeguarding before classroom use.

Guide generation does not grade pupils, decide admissions or progression, or assess individual learners. Automated service checks do decide whether a request can proceed and whether a draft passes review. A second AI model is not human intervention. You can ask a person to review a refusal or flag by contacting hello@ailitkit.com or safeguarding@ailitkit.com; we aim to respond within five working days. A content flag is not a finding of misconduct by a teacher.

Do not enter identifiable pupil information, assessment records, safeguarding records, EHCP/SEN documents or staff personal data. AILitKit needs curriculum content only. Warnings and input cleaning do not guarantee anonymisation: information entered accidentally can still be processed and appear in a saved guide or audit evidence. Contact hello@ailitkit.com promptly if this happens.

Acceptable use policy alignment

AI literacy activities should be delivered in line with your school's acceptable use policy and online safety procedures. Where a guide names an external AI tool, AILitKit applies a tool-vetting tier (recommended, conditional, or not approved at this key stage) based on the tool's published age policy, default privacy settings, and educational fit. Schools should still check tools against their own AUP and procurement processes.

Adaptations for every learner

Every guide includes optional notes that open up the activity for the full range of learners in a class: structured entry points, choices of representation, scaffolds, stretch options, and prompts for pupils who already hold strong prior knowledge. These are starting points for any teacher to adapt, not a separate track for a sub-group of pupils.

UK: KCSIE 2026 alignment

AILitKit aligns with Keeping Children Safe in Education 2026, the statutory guidance for schools and colleges in England in force from 1 September 2026. The 2026 edition brings AI-specific provisions into the statutory text, and AILitKit treats AI-generated imagery and AI-simulated interaction as in scope for safeguarding-aware framing accordingly. Specific alignments:

• AI-generated imagery and deepfakes. KCSIE 2026 confirms that AI-generated imagery falls under child-on-child abuse where used to harass, sexualise, or intimidate. AILitKit guides addressing online harms surface this in age-appropriate ways and signpost CEOP for reporting.
• Preventive education. AILitKit supports schools in delivering preventive education on online harms, deepfakes, and AI-simulated interaction through discussion and debate activities that build critical evaluation skills.
• Cybersecurity as a safeguarding matter. Compromised child data is a safeguarding concern, not only an IT one. AILitKit does not require student personal data and prohibits its submission, and the in-product tool-vetting tier flags any third-party tool we name against its data-protection posture.
• Annual filtering and monitoring review (KCSIE Part 2). Governing bodies must review filtering and monitoring effectiveness at least annually. AILitKit's Whole-Curriculum guides and the per-org safeguarding-decision activity view can support evidence for these reviews; the safeguarding-decision view is intended to surface patterns to your DSL in the same way a filtering-and-monitoring solution does.

For the full statutory text, see the DfE's Keeping Children Safe in Education 2026 on gov.uk.

UK: DfE Generative AI Product Safety Standards (January 2026)

AILitKit aligns with the DfE Generative AI Product Safety Standards, published in January 2026, including the five named risks the standards call out:

• Cognitive offloading. Every guide is positioned as a planning aid, not a substitute for the teacher's professional judgement; the in-product copy and the printed footer reinforce teacher review.
• Anthropomorphism. The product surfaces itself as a tool, not a colleague or assistant. There is no persona, no first-person voice, and no claim of authorship.
• Manipulation. The classifier sits in front of every generation and the prompt instructs the model to use age-appropriate, awareness-not-instruction framing on safeguarding-sensitive topics.
• Emotional dependence. AILitKit does not provide companionship or open-ended conversation; it accepts a curriculum context and returns a guide.
• Distress detection. The product is teacher-only and does not interact with pupils, so the standard's pupil-distress-detection provisions do not engage. Free-text requests are not monitored as an emergency or distress service. Contact the safeguarding inbox directly for a product concern, and use your school or emergency support route for urgent personal concerns.

AILitKit also references the DfE's Generative Artificial Intelligence in Education (2025) and the Filtering and Monitoring Standards where relevant.

EU: AI Act and responsible AI

AILitKit supports EU schools in meeting the transparency and literacy obligations of the EU AI Act:

• AI literacy (Article 4). The Act requires providers and deployers of AI systems to ensure sufficient AI literacy among staff. AILitKit directly supports this by helping teachers build AI literacy into their practice and document staff CPD through the in-product Responsible AI for Teachers course.
• Transparency (Article 50). Every AILitKit guide is clearly labelled as AI-generated. Teachers are reminded in-product and contractually that they must review the guide before classroom use.
• Intended use. Guide generation supports teacher planning; it does not grade pupils, decide admissions or progression, or proctor exams. Automated content checks are described above. Regulatory assessment depends on the actual purpose and deployment; this statement is not a certification.

AILitKit also aligns with DigComp 2.2 and its updated AI-specific competencies.

US: responsible AI in schools

For US schools, AILitKit supports responsible AI use alongside existing safeguarding standards:

• ISTE Standards. Activities align with the ISTE Digital Citizen and Computational Thinker competencies.
• AI4K12 Five Big Ideas. Guides can be mapped to Perception, Representation and Reasoning, Learning, Natural Interaction, and Societal Impact.
• District policies. Activities should be delivered in accordance with your district's acceptable use policy and any state-specific AI guidance.
• No student data. Teachers must not submit student personal data; accidental disclosure remains possible and must be reported. See our US Privacy and Compliance page.

UAE: MoE, KHDA, ADEK, and Wadeema's Law alignment

AILitKit supports UAE schools in meeting the Ministry of Education AI literacy mandates effective from the 2025-26 academic year, and in operating within the relevant regional regulators:

• MoE AI Curriculum. The UAE mandates AI literacy integration across all K-12 subjects. AILitKit generates guides mapped to the MoE's seven AI domains.
• KHDA AI Literacy (Dubai). Dubai private schools under KHDA jurisdiction can use AILitKit guides aligned to the KHDA AI Literacy Framework, launched February 2026.
• ADEK (Abu Dhabi). Schools under ADEK jurisdiction receive guides framed against ADEK's curriculum and safeguarding expectations.
• Wadeema's Law (Federal Law No. 3 of 2016). Safeguarding framings in AILitKit guides reference Wadeema's Law and the UAE National Child Protection Policy in place of UK statutory references when the school is operating in the UAE.
• Teacher-led approach. Activities are designed for teacher-led delivery, in line with MoE guidance that AI tools in schools should be supervised by qualified educators.
• Data protection. AILitKit's data practices align with UAE PDPL requirements. See our UAE Data Protection page.

Review cadence

This statement is reviewed at least every six months and immediately after any change to the safeguarding gate, the classifier model, the admin-visibility surface, or the statutory frameworks named above. The “Last updated” date at the top of this page records the most recent revision.