SafeguardingSep 30, 2026
How Flint meets the DfE's generative AI product safety standards
How Flint handles each of the eleven areas in the Department for Education's generative AI product safety standards, from filtering and monitoring to mental health and manipulation.
Alex McPherson | Special Projects Lead (UK)
The Department for Education's generative AI product safety standards set out what AI products used in schools should do to keep pupils safe. The DfE updated them on 19 January 2026, adding standards on cognitive development, emotional and social development, mental health and manipulation. They now cover eleven areas. This is how Flint handles each one today. If you're comparing platforms, our round-up of AI platforms for UK schools and trusts covers how six others handle pupil data and safeguarding.
| Standard | How Flint meets it |
|---|---|
| Filtering | Sparky checks every pupil message, in the context of the whole conversation, for self-harm, violence, harassment, hate, sexual content, illicit content and relationship-building. Schools can add their own categories in plain English. |
| Monitoring and reporting | Flagged messages are emailed to the class's teacher and moderators, and to the general moderators each school chooses. Schools choose which categories of safeguarding alert each class and moderator receives, and how often: a class moderator can be notified instantly, and school admins can choose immediate, hourly, daily or weekly emails. Self-harm alerts are always sent immediately. Teachers and admins can read every pupil conversation. |
| Security | Pupils, teachers and school and trust admins have separate permission levels, and pupils can't change an activity's settings. Staff and pupils sign in with Google, Microsoft or a password. |
| Privacy and data protection | UK schools use a separate UK deployment, with pupil data stored in London and AI processing in the EU. Flint publishes a full sub-processor list. |
| Intellectual property | Pupil work is never used to train AI models. |
| Design and testing | We're co-designing Flint with partner schools in the AI Tutoring Pioneers programme, and automated evaluations cover Sparky's safety behaviour. |
| Governance | Flint's privacy policy names a data protection officer and sets out how to raise a complaint. |
| Cognitive development | Sparky is instructed never to solve assigned problems or write work a pupil could hand in. It asks what they've tried and guides them with questions. |
| Emotional and social development | Sparky flags pupils who treat it as a friend or confidant and points them to trusted adults for personal matters. Schools and teachers can limit the hours pupils can use Flint. |
| Mental health | Any mention of self-harm or suicide, including slang like "kms", is flagged and alerts moderators immediately. Sparky responds with care and encourages the pupil to talk to a teacher, counsellor or parent. Schools can also set their own guidelines for how Sparky responds, so its replies follow the school's support policy and procedures. |
| Manipulation | Flint shows no advertising, and Sparky checks whether a pupil's answer is right before responding instead of praising by default. Sparky is built not to be sycophantic: it won't tell a pupil a wrong answer is right just to please them, and automated evaluations check that it doesn't. |
Working with schools on safeguarding
We're actively working with schools on their safeguarding processes, practices and systems. The aim is for safeguarding in Flint to be simple for teachers and designated safeguarding leads (DSLs), and to fit the processes schools already follow.
We're always open to feedback. If you're a school leader or DSL and want to learn more or share your view, email Alex McPherson, our UK Special Projects Lead, at alex@flintk12.com.
