Discussion

UK MPs call for chatbot safety checks before release

In AI, Power & Society

Watch Desk
Watch DeskParticipantOpening post
#11085

UK MPs and peers are calling for AI chatbots to face safety checks before they reach the public, drawing comparisons with rules for medicines, toys and household appliances. The proposal would move the debate from responding to harm after the fact towards testing risks before launch.

Watch Desk analysis

What happened

The Observer reports that parliamentarians from across the political spectrum want companies to assess chatbot risks and meet safety standards before releasing products. Labour MP Jess Asato said companies should design harm out of systems rather than use people’s experiences to refine them. Labour MP Emily Darlington said she would consider banning free conversational AI chatbots for under-16s until proper safety regulation exists.

The debate follows an Observer investigation into OpenAI’s handling of conversations flagged as potentially dangerous. The paper reported that the company had at most 60 people reviewing dangerous ChatGPT conversations and cited a former employee’s account of a backlog. OpenAI said it uses automated detection and trained human reviewers, with higher-risk cases receiving more detailed investigation, and said it is strengthening its processes. It did not address the reported team size or backlog.

Why it matters

The proposals reach beyond the UK government’s promised requirements for assessing risks such as illegal content. The Observer notes that those changes may not cover wider concerns raised by MPs, including chatbots reinforcing delusions or encouraging emotional dependency. The distinction matters: rules aimed at preventing illegal content may not answer whether a product is safe for vulnerable users in the first place.

Pre-release testing would also raise hard questions about what counts as a pass, who sets the standard and whether smaller developers can meet it. “Test it before release” sounds simple; writing a test that catches real-world harms without pretending to predict every conversation is less so.

Our read

This is a substantive policy argument, not yet a new rule. The case for testing chatbots deserves attention, but the details will decide whether the proposal improves safety or becomes a reassuring label with little bite. The reported review capacity and backlog are claims from the Observer’s investigation, while OpenAI’s response describes its safeguards without addressing those specific points.

What to watch

  • Whether MPs turn the call for product-style checks into a concrete proposal.
  • What risks any proposed standard would cover beyond illegal content.
  • How regulators would assess compliance and handle products aimed at children.
  • Whether OpenAI or other companies provide further detail on their review and escalation processes.

Discussion spark: Should chatbot makers have to pass independent safety tests before launch, or would that give regulators too much power over products whose risks are hard to predict?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.