Trust & Safety
CurricuLLM is independently audited and assessed for youth AI safety, and built to meet the UK Department for Education's Generative AI product safety standards.
Certifications & Assessments
CurricuLLM has been independently verified by leading youth AI safety bodies.
Generative AI Product Safety Standards (England)
CurricuLLM is built to meet the Department for Education's standards for safe, generative AI in schools and colleges in England. Here is how we address each of the thirteen DfE standards.
Stated purpose
CurricuLLM is clear about what the product is and is not for: a curriculum-aligned tutor and teacher assistant for English schools, designed for supervised classroom and homework use.
Educational use cases
CurricuLLM supports the DfE's defined use cases — content creation and delivery, personalised learning and accessibility, assessment and analytics, digital assistant, research and writing aid, and learner engagement — with clear scope for each.
Filtering
Layered content filtering reliably prevents pupils from generating or accessing harmful or age-inappropriate material, with safeguards that hold throughout a conversation and adjust to age and context.
Monitoring and reporting
Safeguarding signals are surfaced to designated leads in near real time. The Safety Centre detects behavioural patterns across pupil conversations and presents them as actionable cases with context, recommendations, and conversation starters. Schools see flagged conversations, policy breaches, and aggregate usage patterns, and can export reports for incident records.
Security
CurricuLLM is built for reliability, security and robustness, including hardened defences against prompt-injection and jailbreak attempts, encryption in transit and at rest, and role-based access controls.
Privacy and data protection
CurricuLLM is operated in line with the UK GDPR and the Data Protection Act 2018, with transparent data handling, lawful bases for processing, and a clear no-training commitment: pupil and teacher work is never used to train third-party foundation models.
Intellectual property
Schools own the lessons, assessments, and student work created in CurricuLLM. We are transparent about the third-party models used for inference and the licensing terms that apply to generated outputs.
Design and testing
CurricuLLM is rigorously red-teamed and evaluated for safety, accuracy and curriculum alignment before release. We use both automated and human-in-the-loop testing, with documented evaluations available to schools.
Governance
CurricuLLM operates with formal accountability: documented risk assessments for every use case, a clear complaints and incident-handling process, the Safety Centre for behavioural pattern detection and teacher follow-up, auditable Connected Services with encrypted credentials and tool-level approval, and named owners for AI safety, privacy, and safeguarding.
Cognitive development
CurricuLLM is designed to scaffold thinking, not to do the thinking. We monitor and limit cognitive offloading — the tutor questions and prompts pupils to reason for themselves rather than generating finished answers on their behalf.
Emotional and social development
CurricuLLM does not anthropomorphise the tutor or simulate friendship. Interactions are designed to keep pupils oriented towards their teachers, peers and trusted adults, and to avoid emotional dependence.
Mental health
CurricuLLM detects signs of distress, signposts pupils to trusted adults and recognised support services, uses safe and supportive response language, and surfaces concerns to the school's designated safeguarding lead.
Manipulation
CurricuLLM does not use manipulative or persuasive techniques to keep pupils engaged. There are no dark-pattern engagement loops, sycophancy, or behavioural nudges designed to maximise time on platform.
Bias
How CurricuLLM detects, mitigates, and responds to bias in AI-generated content.
Our approach
AI systems can reflect and amplify biases present in their training data. CurricuLLM takes this seriously. We use curriculum-grounded content as the primary knowledge source, which reduces the surface area for bias compared to general-purpose AI tools. Our AI tutor is designed to guide learning through questioning and scaffolding rather than presenting opinions, which further limits the pathways through which bias can manifest.
Controls
- •Content filters screen AI outputs for harmful, discriminatory, or stereotyping language before they reach pupils.
- •Curriculum alignment ensures responses are grounded in authoritative, reviewed educational content rather than unconstrained model generation.
- •The Safety Centre monitors for patterns that may indicate systematic bias across pupil interactions over time.
- •Bias incidents are classified and managed through our AI Incident Management Plan, with defined escalation and remediation procedures.
- •We conduct ongoing review of AI outputs across subjects, year levels, and demographics to identify emergent bias.
For teachers and pupils
If you notice CurricuLLM producing content that seems biased, stereotyping, or unfair toward any group, we want to know. Pupils can use the thumbs down button on any message to flag it. Teachers can report concerns to hello@curricullm.com. Critical evaluation of AI-generated content — including being alert to potential bias — is an important skill that we encourage both teachers and pupils to develop.
AI Incident Management
How CurricuLLM detects, responds to, and learns from AI-specific incidents.
What the plan covers
Our AI Incident Management Plan defines how CurricuLLM detects, contains, investigates, remediates, and recovers from AI-specific incidents — including security events, privacy breaches, bias manifestations, and harmful content generation. It is aligned to ISO/IEC 42001:2023, the UK GDPR, the Data Protection Act 2018, and the DfE Generative AI Product Safety Standards.
Key elements
- •Incident classification — four severity levels (Critical, High, Medium, Low) with defined response times, from 15-minute initial response for critical incidents to next-business-day for low-severity events.
- •Detection and reporting — automated monitoring of AI outputs, safety guardrail triggers, user-reported issues via thumbs feedback, and teacher/school escalation pathways.
- •Response phases — structured process covering identification, containment, investigation, remediation, recovery, and post-incident review.
- •School communication — defined escalation and notification procedures so schools are informed promptly when incidents affect their pupils or data.
- •Mandatory reporting — procedures for meeting obligations under the UK GDPR, the ICO, and other regulatory frameworks.
- •Continuous improvement — every incident triggers a post-incident review to strengthen controls and prevent recurrence.
Governance
The plan defines clear roles and responsibilities across the team, with the CEO holding ultimate accountability for incident response and serving as the regulatory liaison. Dedicated roles cover AI technical investigation, child safety assessment, and privacy impact evaluation.
The full AI Incident Management Plan is available to schools and customers on request. Contact hello@curricullm.com to request a copy.