AI safety moves into public hearings
A UK inquiry opens as Europe seeks incident evidence. Scrutiny matters; hearings and information requests are not completed safeguards.
AI safety is moving into public hearings and requests for evidence. The useful question is what those processes will actually deliver: independent access, explanations people can inspect, and decisions that change how powerful systems are developed.
On October 14, a UK parliamentary committee announced a continuing inquiry into AI adoption, safety and governance. Its first hearing took place on October 13, with former OpenAI and Anthropic employees, representatives of four major developers, and government officials. The committee plans to revisit commitments and unresolved questions. It says targeted calls for evidence will follow; this is not an invitation to submit anything immediately, nor a new safety law.
What is confirmed
The European Commission's October 12 account of its scientific panel meeting says the AI Office has received serious-incident reports and requested information about developers' security practices. Further questions are being prepared. The panel recommends stronger scrutiny of models used internally during development and testing, better incident-investigation capacity, and international cooperation.
That official update does not identify the developers or announce a penalty. Euronews identifies OpenAI and Anthropic in today's reporting, while describing disagreement over how the AI Act applies to testing incidents. An information request is not a finding that a company broke the law. We should follow the evidence without turning a reported investigation into a verdict.
Britain's October 12 announcement also confirms that its AI Security Institute is evaluating the latest models following work with developers and the US government. Ministers are considering stronger protections, incident response, and near-miss reporting through the Cyber Security and Resilience Bill regime. Those details are still being developed. Model access is useful, but the announcement alone does not establish that evaluation can stop an unsafe release.
What to watch next
Our editorial test is concrete: can outside evaluators inspect the relevant systems, can they explain significant findings publicly, and can authorities require a response? A hearing can expose a problem; a report can document it. Neither automatically fixes it.
For readers who want to participate, follow the committee's specific evidence requests when they appear. Ask representatives for clear responsibilities and enforceable safeguards, rather than treating every safety announcement as an accomplished result. Our Take action guide offers other routes, including the superintelligence statement we support. Enjoying ChatGPT and asking who controls its future belong in the same conversation.