England's schools got a position paper, then rules for vendors

The classroom note

A school considering a generative AI tool for lesson planning in spring 2023 had one government page to check first. On 29 March 2023 the Department for Education published its position on generative artificial intelligence in education. Two years later a second document arrived, addressed not to schools but to the companies selling into them: product safety standards for edtech developers and suppliers. Read together, they show a department splitting the problem between a school's choices and a vendor's product design.

What the evidence says

The position statement mixes three kinds of language deliberately. Personal data "must be protected in accordance with data protection legislation" — an existing duty, not new to AI. It is separately "recommended that personal data is not used in generative AI tools" at all, beyond what the law demands. Schools, colleges and awarding bodies "need to continue taking reasonable steps... to prevent malpractice," and safety "should be the top priority when deciding whether to use" it. The document also says AI "has demonstrated" it can cut administrative workload — a claim to note, not a measured finding, since no study accompanies it. The safety standards, last updated 19 January 2026, set what they call a "minimum expected" for products; their added manipulation category since first publication shows the standard has been revised, not fixed.

The implementation question

Two standards translate into concrete behaviour. "Progressive disclosure" means a tool should withhold a final answer and offer hints first — a pedagogy choice a vendor's design can quietly override. A ban on anthropomorphising means a chatbot should not imply personhood and should redirect a struggling student toward "a classmate or teacher" rather than deepen reliance on itself. A mental-health expectation asks products to detect distress signals and follow "soft signposting" or a safeguarding flag — a new, software-mediated entry into a school's safeguarding duty that the standard does not say how to route to a named lead.

What holds and what fails

Both documents are guidance and expectations, not statute: the binding duties are the pre-existing data protection and safeguarding law they restate, not the AI-specific advice on top. That holds as a procurement checklist — a school can put the safety standards to a vendor as questions, though nothing compels an answer. It fails where keeping current with two separately updated documents sits entirely with the school, and where a safeguarding flag has no specified school-side owner. This is an editorial reading: splitting advice to schools from standards for suppliers moves the safety question upstream without making it enforceable there either.

• Has the vendor evidenced progressive disclosure and non-anthropomorphising design, or only asserted the product is "safe for education"?
• If the product raises a safeguarding flag, which named person at the school receives it, and on what timescale?
• When was the position statement and the safety standards document each last checked against the version currently in force?

Neither document requires a school to act; both describe what "good" looks like from a particular vantage point, and a reader should notice which one — school or vendor — before treating either as settled.