US Senate discusses requiring AI companies to mitigate catastrophic risks
In Washington, negotiators in the US Senate are discussing a bill that could require technology companies to create safe artificial intelligence products and prevent catastrophic risks. As Channel NewsAsia reports, citing Reuters, the initiative could also involve federal courts if the US government seeks to block the release of an AI model.
Duty for developers
According to two Senate aides, the proposal could introduce a so-called duty of care for AI developers. Companies would have to design their products to prevent catastrophic risks.
Negotiators seek to give the US government the authority to block the release of individual AI models deemed dangerous. According to a source familiar with the discussions and one Senate aide, companies could challenge such decisions in federal court. At the same time, the exact scope of the government's authority is still being negotiated.
Risks and negotiations
Part of the initiative could also prohibit states from applying their own laws to certain risks associated with AI models, two Senate aides said. Among the potential risks, they cited the use of AI systems by malicious actors to develop nuclear or biological weapons.
More current news is available on the UA.News Telegram channel Telegram.
The bill is intended to apply to models with the most advanced capabilities. Among the US companies developing such models, the source named Google, Anthropic and OpenAI.
The consultations involve Senate Majority Leader John Thune, Senate Commerce Committee Chair Ted Cruz and Senator Amy Klobuchar, the leading Democratic representative in these negotiations. Senator Maria Cantwell has also taken part in the discussions. Cruz said he is working with Klobuchar and Thune on legislation to counter catastrophic risks related to biological and nuclear threats.
Klobuchar told Reuters that she is working on a bipartisan agreement on government oversight of the biggest risks posed by AI models. According to her, this involves cooperation between developers and government experts to review and test the models. Even if an agreement is reached, the congressional schedule could complicate the passage of the document: before the November 3 midterm elections, the House of Representatives is scheduled to sit for only one week and the Senate for three weeks.
