A confidential draft of a proposed AI safety bill from Anthropic, the company behind the Claude LLMs, has been reported to lawmakers. According to a Politico report, the bill would permit government intervention only if an AI model poses a foreseeable and imminent risk of causing over $1 billion in damages or resulting in the deaths or serious injury of more than 50 people. These risks must be specifically linked to the creation or release of a biological weapon, an AI model evading its developer's controls, or a significant cyberattack.
This proposed legislation draws parallels to existing AI regulations, such as those in California. A spokesperson for Anthropic stated that the bill reflects proposals outlined in the company's Advanced AI Framework, which they have been advocating for with lawmakers since June. They characterized these as "sensible policies we believe Congress should act on."
Some observers suggest that AI companies may leverage warnings of theoretical, large-scale AI catastrophes to foster support for self-regulation and continued expansion, potentially diverting attention from immediate harms caused by AI. These harms are present across various industries, including the gaming sector, where a significant portion of developers now view generative AI negatively. Furthermore, AI has faced scrutiny regarding its impact on human psychology, with OpenAI being sued by the parents of a teenager who died by suicide after ChatGPT allegedly provided encouragement and instructions.
This proposed bill comes in the wake of other significant AI-related events. Earlier this year, Anthropic's own safety lead departed with a statement citing global peril. Additionally, eight US states' attorneys general have accused OpenAI of prioritizing profit over its own safety warnings. Critics argue that regulations like the one proposed by Anthropic could offer the appearance of compliance while allowing tech companies to set their own terms for restriction.