OpenAI and Anthropic Tell British Lawmakers: Regulate Us
OpenAI and Anthropic told a cross-party committee of British MPs that some binding rules are needed for frontier AI, arguing voluntary commitments are not enough. Their testimony highlighted a split among major AI firms and renewed questions about whether the UK can manage risks such as AI-enabled bioweapons without stronger powers.
Anthropic backed binding rules, while OpenAI supported mandatory measures such as incident reporting and an international convention. Google DeepMind favoured focusing on standards, and Meta pointed to existing regulators. Officials said the UK’s AI Security Institute tests models only when companies provide access and cannot stop a release it considers dangerous; OpenAI said UK access to its pre-release models depends on what the US government permits.
- OpenAI and Anthropic backed binding AI rules.
- Google DeepMind and Meta took different approaches.
- UK safety testing relies on voluntary company access.
New here? Start with this
Frontier AI refers to the most advanced general-purpose AI systems, which can handle a wide range of tasks and may become more capable over time. Companies including OpenAI, Anthropic, Google DeepMind and Meta develop these systems, while governments are weighing how to oversee them.
The UK’s AI Security Institute can test models when companies give it access, but it cannot prevent a company from releasing a system it considers dangerous. Some risks under discussion include AI being used to help create biological weapons, as well as failures that could cause wider harm.
British lawmakers are considering whether existing regulators and voluntary company promises are enough, or whether new binding rules are needed. The debate also involves access to unreleased systems, which may depend on decisions by companies and, in some cases, the US government.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Binding rules could give the UK clear powers to require safety testing, incident reporting and safeguards before highly capable systems are released. Supporters argue that voluntary promises may be uneven or hard to enforce, while risks such as AI-enabled bioweapons could affect the public beyond the companies developing the models.
The case against
A more limited, standards-based approach could avoid imposing rules before the technology and its risks are well understood, which might otherwise slow useful innovation or entrench firms able to absorb compliance costs. Advocates may also argue that existing regulators and international coordination are better suited to oversight, especially when UK authorities cannot always access models developed or controlled abroad.