Anthropy has arrived against a proposed Illinois law backed by OpenAI that would protect AI labs from liability if their systems are used to cause large-scale harm, such as mass casualties or more than $1 billion in property damage.
The SB 3444 dispute creates fresh lines of contention between Anthropic and OpenAI over how to regulate AI technology. While artificial intelligence policy experts say the legislation has little chance of becoming law, it has nonetheless exposed political divisions between two leading U.S. artificial intelligence labs that may become more critical as rival companies raise their lobbying activities across the country.
Behind the scenes, Anthropic has been lobbying state Sen. Bill Cunningham, sponsor of SB 3444, and other Illinois lawmakers to either make major changes to the bill or eliminate it in its current form, according to people familiar with the matter. In an email to WIRED, an Anthropic spokesperson confirmed the company’s opposition to SB 3444 and said it had had promising conversations with Cunningham about using the bill as a starting point for future AI legislation.
“We oppose this bill. Good transparency legislation must ensure public safety and accountability for the companies developing this advanced technology, not provide a get-out-of-jail-free card for any liability,” Cesar Fernandez, head of state and local government relations at Anthropic in the US, said in a statement. “We know Senator Cunningham cares deeply about AI security, and we look forward to working with him on changes that would instead combine transparency with real accountability for mitigating the most serious harms that border AI systems can cause.”
Representatives for Cunningham did not respond to a request for comment. A spokesman for Illinois Governor J.B. Pritzker sent the following statement: “While the Governor’s Office will monitor and review many of the artificial intelligence bills pending in the General Assembly, Governor Pritzker does not believe that large technology companies should ever be given a full shield to avoid the responsibilities they should have to protect the public interest.”
At the heart of OpenAI and Anthropic’s dispute over SB 3444 comes down to who should be liable in the event of an AI-based disaster – a potential nightmare scenario that U.S. lawmakers have only recently begun to grapple with. If SB 3444 were passed, an AI lab would not be liable if a bad actor used its AI model, for example, to create a biological weapon that kills hundreds of people, provided the lab developed its own security framework and posted it on its website.
OpenAI argues that SB 3444 reduces the risk of solemn harm from pioneering artificial intelligence systems by “still allowing this technology to get into the hands of people and businesses – small and large – in Illinois.”
The creator of ChatGPT says it has worked with states like Modern York and California to create a so-called “harmonized” approach to regulating artificial intelligence. “In the absence of federal action, we will continue to work with states – including Illinois – to develop a consistent security framework,” OpenAI spokeswoman Liz Bourgeois said in a statement. “We hope this state legislation will influence a national framework that will help ensure the United States continues to lead.”
On the other hand, Anthropic argues that companies developing pioneering AI models should be at least partially responsible if their technology is used for widespread social harm.
Some experts say the bill would eliminate existing regulations designed to deter companies from misbehaving. “Liability already exists at common law and provides a strong incentive for AI companies to take reasonable steps to prevent foreseeable threats from AI systems,” says Thomas Woodside, co-founder and senior policy advisor at the Secure AI Project, a nonprofit that has helped develop and advocate for AI security regulations in California and Modern York. “SB 3444 would take the extreme step of almost completely eliminating liability for serious damages. Diluting liability, which in most states is the most important and already existing form of legal liability for AI companies, is a bad idea.”