Key information
A confidential draft of an AI safety bill from Anthropic has been presented to US lawmakers. According to reports, the proposal suggests that government intervention should only occur if an AI model presents a foreseeable, imminent risk of causing over $1 billion in damages or resulting in the death or serious injury of more than 50 individuals.
The draft specifies that these high-stakes thresholds apply specifically to risks originating from cyberattacks, the development of biological weapons, or instances where a model successfully bypasses its developer's internal controls.
An Anthropic spokesperson stated that these measures align with the company's Advanced AI Framework, which they have been promoting to state and federal officials since June. The company characterizes these as sensible policies that Congress should adopt.
Critics argue that such regulations may allow tech companies to appear compliant while effectively restricting operations only on their own terms. Some observers suggest these proposals leverage fears of catastrophic outcomes to facilitate self-regulation, potentially ignoring existing harms caused by current AI technologies.
