OpenAI and Anthropic Tell British Lawmakers: Regulate Us

Two of the world's biggest artificial intelligence companies, OpenAI and Anthropic, told British lawmakers on Tuesday that the government needs to pass new laws to keep AI safe. The companies said that letting AI firms make their own voluntary promises is no longer good enough. The hearing took place before a group of elected members of the British Parliament called the Business, Innovation, Science, and Trade Committee.

Anthropic's head of Europe, Liam Booth-Smith, told the committee that the rules should be legally binding. He said his company has consistently believed that voluntary commitments are not enough and that real laws are needed. Binding rules are laws that companies must follow or face real consequences.

OpenAI's vice president Tom Duff Gordon also supported new laws, though he was a little more careful about it. He said some things, like reporting when AI systems cause problems, should be required by law. 'We don't think basing everything on voluntary commitments for frontier risk is a good long-term plan,' he said.

However, two other big AI companies did not agree. Google DeepMind and Meta both said they did not support new laws for AI safety. Google DeepMind's Owen Larter said the focus should be on setting safety standards, while Meta's Rob Sherman pointed to rules that already exist in Britain.

The chair of the committee, Liam Byrne, said the hearing was meant to show that AI companies have very different views on safety. He wanted to make it clear that the government cannot simply trust AI companies to police themselves. He said the goal was to prove that voluntary self-regulation is not reliable.

On the same day as the hearing, the director of Britain's MI5 spy agency, Ken McCallum, gave a separate warning about AI dangers. He said one of his biggest worries is that AI could be used to help people create biological weapons. While this has not happened yet, he said the risk is serious enough to take very seriously.

Britain has a government body called the AI Security Institute, or AISI, which tests powerful AI models before they are released to the public. But it can only test a model if the company that built it agrees to share it. Henry de Zoete, the head of AISI, admitted that his organization has no power to stop a company from releasing a model it believes is too dangerous.

Four former workers from AI companies also spoke at the hearing. One of them, Daniel Kokotajlo, who used to work at OpenAI, told lawmakers: 'I don't think the companies can be trusted to self-regulate.' Another former OpenAI worker said he was not confident that the AI industry as a whole has a strong safety culture.

After the hearing, committee chair Byrne said he wants Britain to use its upcoming presidency of the G20 group of nations to push for an international agreement on AI safety. He said this agreement should not depend on the current leadership in the United States. He believes global cooperation is the best way to make sure AI development stays safe for everyone.

'The risks,' he warned, 'are multiplying faster than our safety regime can control.' This means that dangers from AI are growing more quickly than the rules meant to stop them. Byrne and others are pushing governments around the world to act before it is too late.