Anthropic proposes a transparency framework for frontier AI developers

Anthropic published a proposal for a minimum federal transparency standard aimed specifically at the largest frontier AI developers — those with roughly $100 million in annual revenue or around $1 billion in annual research and development spending — while explicitly exempting smaller companies and startups from any new obligations.

The framework asked covered developers to document, in a public “secure development framework,” how they assess and mitigate risks including chemical, biological, radiological and nuclear misuse and the risk of models acting autonomously against their operators’ intent; to publish that framework on their own websites, with redactions permitted for genuinely sensitive material, alongside a self-certification of compliance; and to release system cards at deployment describing testing procedures, evaluation results and mitigations, updated whenever a model changed substantially. It also proposed that laws prohibiting false statements about compliance should extend existing whistleblower protections to employees who raised safety concerns.

Anthropic argued the requirements should stay deliberately lightweight and adaptable rather than prescriptive, on the reasoning that rigid rules would age quickly against a fast-moving field, and that the goal was baseline public accountability rather than a licensing regime. The proposal arrived the same day the European Commission published its own General-Purpose AI Code of Practice, a voluntary compliance mechanism under the EU AI Act covering similar ground on transparency and safety disclosure — giving frontier labs, within a single week, competing templates for what “transparency” should mean in binding US law versus existing EU regulation.