Regulation of Artificial Intelligence Industry Discussed at British Parliamentary Committee Meeting

A high-stakes meeting between top artificial intelligence firms and British lawmakers has shed light on the industry's stance on regulation. The Business, Innovation, Science, and Trade Committee held an evidence session, bringing together four AI companies, including OpenAI and Anthropic, along with former employees turned whistleblowers and senior government officials.

Anthropic's head of Europe, Liam Booth-Smith, made a clear call for binding regulations, stating that the company believes AI safety requires a statutory framework. In contrast, OpenAI's vice president, Tom Duff Gordon, was more measured in his approach, advocating for a statutory framework but emphasizing the need for mandatory elements, starting with incident reporting.

Duff Gordon expressed concerns about relying solely on voluntary commitments to address frontier risks, suggesting that this approach may not be sustainable in the long term. This apparent divide within the AI industry has been put on record by the British parliamentary committee, highlighting the conflicting views of leading companies on how best to regulate their activities.

The hearing highlighted the differing views within the AI industry regarding regulation, a topic discussed by a British parliamentary committee earlier this year.

Meanwhile, just a few hundred meters away from where these executives were giving evidence, Britain's security service director was issuing a stark warning about the dangers of AI misuse.

MI5 director Ken McCallum used his annual threat update to highlight bioweapon misuse as a significant threat stemming from AI capabilities.

McCallum emphasized that the misuse case he considered most pressing is the potential for AI to alter the stakes in existing or new bioweapons, and this risk has yet to be realized.

The UK's current system governing AI security has been revealed to have significant limitations, with civil servants acknowledging how little pressure it applies on companies.

The Cabinet Office's Cat Little confirmed that the current AI security regime in the UK is voluntary, with companies free to operate as they see fit without facing significant consequences.

When pressed on what would happen if a company released an AI model deemed too dangerous by AISI, the organization responsible for overseeing AI security, de Zoete acknowledged that AISI has no power to prevent such releases. This lack of enforcement underscores the limitations of the current system, which relies on companies' voluntary compliance with guidelines.

The tension between UK and US regulations was also highlighted during the meeting, as executives from OpenAI and Anthropic made it clear that their access to British authorities is dependent on their relationships with Washington. OpenAI's Duff Gordon noted that they provide their models predeployment for review by CAISI, its US counterpart, while Anthropic's Booth-Smith pointed out that any issues with UK AISI are ultimately a matter for the US government to resolve.

Former researchers from OpenAI and other labs have shared their concerns about the development and deployment of AI technology in the UK. One such researcher, Daniel Kokotajlo, who previously worked at OpenAI, expressed his distrust in the companies' ability to self-regulate. He emphasized that he believes these organizations cannot be relied upon to ensure responsible use of AI.

Geoffrey Irving, another lab researcher, stated that the labs feel they are locked in a competitive race with governments unwilling to intervene. However, Irving disputed this notion, pointing out that it is based on false assumptions. His comments highlighted the complex dynamics at play between private companies and government agencies regulating AI development.

Following the hearing, committee chair Byrne announced plans for his team to thoroughly examine the UK's role in AI governance. He aimed to leverage Britain's upcoming G20 presidency to broker an international convention on AI that would not rely on current US leadership.