AI Firms Split at UK Hearing Over Binding Safety Rules
The debate exposed a key limit: Britain’s AI safety institute can test models, but may lack power to compel access or stop a launch.
OpenAI and Anthropic backed binding AI rules at a UK parliamentary hearing on October 13. Google DeepMind and Meta did not endorse a new statutory framework. The hearing put a practical question before lawmakers: can safety checks be enforced before powerful models reach the public?
The Business, Innovation, Science and Trade Committee questioned company representatives, former lab researchers and government officials about independent testing, access to unreleased models, incident reporting and regulators’ powers. No new law was passed.
The hearing exposed a gap between testing an AI model and having the legal power to compel access or stop its release.
Four companies, different positions
- Anthropic: Head of Europe Liam Booth-Smith said voluntary commitments were not enough and argued for binding rules.
- OpenAI: Vice-president Tom Duff Gordon backed a statutory framework and said some requirements should be mandatory, starting with incident reporting, according to WIRED’s account of the hearing.
- Google DeepMind: Owen Larter focused on standards rather than a new statutory framework.
- Meta: Rob Sherman pointed to Britain’s existing regulators.
The split is over whether safety expectations should remain voluntary or become enforceable duties.
Testing does not guarantee enforcement
The UK AI Security Institute can assess a model when it gets access. But, as the committee discussed, the institute may not have the power to compel that access or halt a release.
That distinction matters: testing can identify risks, but it does not itself prevent a company from launching a model. The hearing raised whether Parliament should give regulators stronger powers. It produced no new law and set no timetable for one.
What lawmakers considered
Possible safeguards discussed included:
- Required independent testing
- Earlier access to models for evaluators
- Mandatory reporting of serious incidents
The companies offered no shared industry blueprint. For people in the UK, there is no new compliance rule yet. The unresolved choice is whether government should rely on company cooperation or give regulators stronger legal powers.
