Politico recently reported on a confidential draft of a proposed AI safety bill from Anthropic, the company behind Claude LLMs (via Tom's Hardware). The bill, which has been pitched to US lawmakers, would let the government crack down on a given AI model "only if it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people. The risk must stem from the creation or release of a biological weapon, the model evading its developer’s controls or a cyberattack."

That article pointed out similarities between these proposals and those used in California's AI legislation, and shares a quote from a spokesperson for the company. "This bill represents the proposals outlined in Anthropic’s Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June," they said, calling "sensible policies we believe Congress should act on."

Notably, it has been posited that AI magnates leverage warnings of theoretical, wide-reaching catastrophe to manufacture consent for their self-regulation and expansion while doing little to address the harm AI is already doing (both broadly within industries like the one this website revolves around and directly with regards to human psychology).

Just earlier this year, Anthropic's safety lead quit with a note that "the world is in peril" —and just last week, OpenAI was accused by eight US states' attorney generals of ignoring its own safety warnings for the sake of profit. Regulations like what Anthropic has proposed here arguably give tech companies the appearance of capitulation while only restricting them on their own terms.

Source link