Anthropic’s Reported Proposal Could Let the U.S. Block AI Models With Court Approval

On October 17, 2026, AFP reported that Anthropic had presented U.S. lawmakers with a proposal that could let the federal government block an AI model with judicial authorization if an imminent risk could cause more than 50 deaths or serious injuries, or more than $1 billion in losses. The figures and proposed authority are AFP’s account of the later proposal, not provisions of Anthropic’s separate June 2026 framework. AFP’s report

What the reported proposal would allow

The reported trigger combines imminence with a threshold for potential harm: more than 50 deaths or serious injuries, or losses exceeding $1 billion. Judicial authorization would be required for the federal government to block a model under the proposal as AFP described it.

Which developers the June framework would cover

Anthropic’s Advanced AI Framework, dated June 2026, proposes rules for developers that meet both a model-training threshold and one of two company-size tests. A developer would have to train models using more than 10²⁵ FLOPs—a measure of computing operations—and either earn more than $500 million a year in AI-derived revenue or spend more than $1 billion a year on AI research and development.

The framework identifies four risk areas: biological weapons, offensive cyber operations, loss of control, and AI-automated research and development that could amplify the other risks. Its definition of catastrophic risk is based on foreseeable, material risks of significant death, injury, or damage.

Reporting and independent evaluation

For covered developers, the June framework proposes public risk reports at least every six months and reporting defined critical safety incidents to a designated agency within 15 days of discovering an incident—or learning facts that support a reasonable belief that one occurred.

It also recommends that covered developers regularly engage at least one qualified independent evaluator, beginning within six months after the proposed regulation takes effect. The evaluator would review the developer’s risk assessment and have access to relevant models and materials. These duties are recommendations in the June framework.

The emphasis on outside scrutiny is part of a broader debate over who should assess powerful AI systems. NeoTeo previously covered Sarah Heck’s call for external oversight; the later proposal’s reported blocking authority is a separate development.

Court oversight in the June framework

The June framework presents court enforcement as one possible design, rather than selecting a single final mechanism. Under one option, a government agency would seek remedies by filing a lawsuit. The framework also discusses provisional remedies for imminent catastrophic risks, subject to judicial review, and expedited review of enforcement decisions.

Those options describe Anthropic’s June recommendations. AFP’s account of the later proposal describes judicial authorization for blocking, but does not establish that the June framework’s specific enforcement procedures apply to it.

Legislative status

AFP reported on October 17, 2026, that Congress had not approved related legislation. Anthropic’s reported proposal and its June framework were proposals, not laws in force.