Anthropic says it wants no get-out-of-jail-free card if Claude is used to harm humans

Anthropic and OpenAI have found themselves on opposing sides of a proposed law in the US state of Illinois. Anthropic has strongly opposed this bill, SB 3444, which would limit the liability imposed on AI labs in cases where their systems are used in mass harm, such as large-scale casualties or property damage.

As per a report from Wired, the Dario Amodei-led firm has been lobbying Illionis lawmakers to change the bill or kill it entirely. Anthropic’s opposition comes just days after OpenAI backed this bill.

Here is what the bill proposes, and why the two companies have different views.

What is the bill?

The bill, named the Artificial Intelligence Safety Act, reduces the liability of AI labs in situations that cause critical harm. It defines 'critical harms' as severe incidents, including the use of AI in creating dangerous weapons or damaging property worth over $1 billion.

That is, for example, an AI company like OpenAI or Anthropic has limited liability if a bad actor uses its tools to develop a bioweapon that kills hundreds of people, as long as the lab drafted its own safety framework and published it on its website.

Why Anthropic opposes the bill?

As per the report, Anthropic has been in touch with the bill’s sponsor, Senator Bill Cunningham, and other state lawmakers. The company wants the lawmakers to make some big changes to the bill or to “kill it as it stands.”

“Good transparency legislation needs to ensure public safety and accountability for the companies developing this powerful technology, not provide a get-out-of-jail-free card against all liability,” Cesar Fernandez, Anthropic’s head of US state and local government relations, said as quoted by Wired.

That is, Anthropic doesn’t want AI labs to get away with limited liability in situations when an AI may be used for severe harm to life or property.

Anthropic has testified in favour of a separate Illinois bill, SB 3261, which would require AI developers like OpenAI and Anthropic to create public safety and child protection plans. These plans would then be tested by third-party auditors to assess their effectiveness.

Why is OpenAI backing it?

OpenAI supports this bill, claiming that it would allow AI technology to be more accessible while reducing the risk of severe harm. OpenAI spokesperson Jamie Radice said, as quoted by Wired, “Reducing the risk of serious harm from the most advanced AI systems while still allowing this technology to get into the hands of the people and businesses.”

Do note that OpenAI has also advocated for a unified federal approach to AI regulation in the US instead, something that is not present at the moment.

OpenAI and Anthropic had different views on Military AI use

Interestingly, the bill comes at a time when OpenAI is working with the US Department of Defense to use its AI tools for the US Military. The Sam Altman-led firm signed a deal with the Pentagon after Anthropic walked away, citing fears that its AI may be used for the development of autonomous weapons or for mass domestic surveillance. OpenAI has insisted that its models cannot be used for such uses either.

Following Anthropic’s exit from the deal, it was designated as a supply chain risk, a rare occurrence for a US firm. Though the company is battling the decision in court.

The US has used AI during its military strikes on Iran. Though the models used were likely made by Anthropic as the Pentagon is still in a transition phase to use OpenAI’s models.