What happened
Anthropic is said to have backed a draft bill that would permit federal intervention – a so‑called “kill switch” – only after an AI system causes at least 50 human deaths or serious injuries, or inflicts $1 billion in damages. The bill would limit the types of harms that count toward the threshold to biological‑weapon attacks, cyber‑attacks, or loss of developer control. The proposal also contains “preemption language” that could override stricter state AI regulations. The information comes from a POLITICO report cited by the activist group More Perfect Union, which posted an Instagram video quoting Anthropic CEO Dario Amodei and highlighting the high‑threshold language.
According to the POLITICO report referenced by More Perfect Union, Anthropic has publicly supported a federal bill that would only trigger a shutdown after an AI system causes at least 50 deaths or serious injuries, or results in $1 billion in damages. The bill’s language limits qualifying harms to three categories: biological‑weapon use, cyber‑attack, or loss of developer control over the model.
The Instagram post from More Perfect Union includes a video of Anthropic CEO Dario Amodei stating, “I would rather they make fun of me than wake up one day and discover that someone used our model, Claude, to kill a bunch of people.” The post also flags “preemption language” that would make the federal bill supersede state AI laws, potentially preventing states from enforcing stricter rules on AI development, deployment, or internal use.
Critics have highlighted the discrepancy between this high‑threshold approach and Anthropic’s earlier public statements emphasizing proactive safety. Commenters on the Instagram post argued that responsibility should lie with the operators and owners of AI systems rather than waiting for mass casualties before federal action.
Source details: thecooldown.com ↗
Why it matters
The high‑threshold approach marks a stark departure from many AI‑safety advocates who argue that federal oversight should intervene well before mass casualties occur. If enacted, the bill could leave a large window for harmful AI deployments to cause damage before any federal response is possible, potentially undermining state‑level protections that are already stricter. The preemption clause could also centralize regulatory authority, limiting states’ ability to tailor rules to local risks such as water use or data‑center impacts. Anthropic’s public backing signals that a leading AI developer is willing to endorse a framework that many in the community consider too lax, which may influence other companies and lawmakers as the bill moves through Congress.
A 50‑death or $1 billion damage trigger creates a large safety gap, allowing potentially dangerous AI systems to operate unchecked for years. This contrasts with proposals from other policymakers that call for earlier, risk‑based interventions.
The preemption clause could erode state‑level AI regulations that address local concerns such as water usage, electricity demand, and environmental impact, concentrating authority at the federal level and possibly slowing more nuanced responses.
Anthropic’s endorsement may lend credibility to the bill among other AI firms, influencing the legislative debate and potentially shaping the regulatory landscape for the entire industry.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').Why can ethical evaluation not be reduced to one model score?
What to watch next
Watch for congressional hearings or amendments that address the 50‑death threshold and preemption language, as well as reactions from state legislators who may push back against federal preemption. Monitor statements from AI‑safety groups and industry peers, which could shape the bill’s final language. Finally, track any formal filing of the bill and whether Anthropic or other firms file comments during the rule‑making process.
Congressional committees may hold hearings on the bill’s thresholds and preemption language; amendments could lower the casualty threshold or carve out exceptions for state laws.
AI‑safety advocacy groups are likely to file formal comments or lawsuits challenging the high‑threshold provisions, which could delay or reshape the bill.
State legislators in tech‑heavy regions (e.g., California, Texas) may introduce counter‑legislation to preserve or strengthen state‑level AI oversight.