Regulation
OpenAI and Anthropic ask for laws that would bind them; Google and Meta don't
Speaking to the British Parliament, OpenAI and Anthropic said voluntary promises are no longer enough. In the same session, the government admitted it cannot stop a model it considers dangerous.
The essentials
4 according to the source · 2 open questions
- According to the sourceAccording to WIRED, Anthropic and OpenAI told British MPs that AI safety laws are needed because voluntary commitments are not enough.1
- According to the sourceGoogle DeepMind and Meta did not back a new law. Google preferred to talk about standards, and Meta pointed out that the UK already has regulators.1
- According to the sourceThe head of Britain's AI safety institute admitted that it only tests the models companies hand over and that it cannot stop a launch.1
- According to the sourceOn the same day, the head of MI5 named the use of AI for biological weapons as the misuse that worries him most, although it has not happened yet.1
Why it matters
Two of the companies building the most advanced models are now publicly asking to be regulated, while two others are pushing back. This suggests that the idea of the industry policing itself is losing support within the industry. The session also made the UK's limits clear: without a law, the UK depends on the companies, and Washington, letting it look.
The details
The session took place on October 13 before a cross-party committee of MPs, according to WIRED1. OpenAI was more cautious than Anthropic. Its vice president Tom Duff Gordon backed a law and an international treaty, and proposed starting by making it mandatory to report incidents1. He also explained that OpenAI gives its models to the equivalent US body before launch, and will keep giving them to the British one as long as Washington allows it1.
Anthropic acknowledged that it could not give the British institute full access to Mythos 5.1 at the time, that it referred the question to the US government, and that the institute now does have access1. Four former researchers from these companies also testified. One of them, Daniel Kokotajlo, said they cannot be trusted to regulate themselves1. The committee chair, Liam Byrne, wants to use the UK's G20 presidency to push for an international treaty1.
What we don't know
- Whether the British government will propose a law, and what obligations it would include.
- Why Anthropic could not give the British institute full access to Mythos 5.1 when it launched.
Background
On October 2 we reported that executives from Google, Anthropic, Meta, OpenAI and Nvidia signed a voluntary safety agreement at the White House, with no law binding them to it. The London session shows that not all of those companies believe that is enough.
Sources
Was this story useful? Yes Not really