AI Firms Sought Restrictions on Safety Evaluators
Leading artificial intelligence companies have requested limits on the independence of safety researchers working within their labs.
Updated on Oct. 13, 2026 in Artificial Intelligence
Live Poll
Should AI companies be forced to share proprietary data for independent safety reviews?
AI companies have requested new restrictions on independent safety evaluators, including the right to review and potentially edit evaluation reports. These firms cite concerns regarding the protection of proprietary information and trade secrets ahead of upcoming initial public offerings.
Why it matters
The tension highlights a fundamental conflict between the need for rigorous, independent AI safety testing and the corporate imperative to protect confidential intellectual property. This struggle over transparency has prompted both congressional action and public pushback from the scientific community.
The current voluntary oversight agreement includes four distinct safety measures for participating labs. Labs are considering requirements such as staff observation of evaluators and prohibitions against evaluators working at multiple competing laboratories.
The players
Donald Trump
Donald Trump is the current President of the United States.
Josh Hawley
Josh Hawley is a United States Senator from Missouri who has introduced legislation regarding AI system stress-testing.
Anthropic
Anthropic is an artificial intelligence research company that has expressed concerns regarding the disclosure of proprietary information to rival firms.
OpenAI
OpenAI is an AI laboratory that currently redacts confidential information from safety evaluation reports.
Microsoft
Microsoft is a multinational technology corporation that anticipates the future establishment of global common rules for AI safety.
The details
Anthropic and OpenAI have expressed concerns that disclosure of technical details to evaluators could inadvertently compromise their competitive advantage. To mitigate this risk, firms have proposed redacting confidential data or requiring approval of final reports before they are published.
Timeline
President Trump and AI executives signed an oversight agreement in September 2026.
Microsoft discussed government regulation of AI on October 12, 2026.
Reports regarding concerns over lab evaluation protocols emerged on October 13, 2026.
The Tech Race
The push by AI labs for editorial control marks a departure from the oversight objectives outlined in the Artificial Intelligence Risk Management and Security Act. This conflict underscores a broader arms race where companies must balance rapid innovation cycles against mandatory safety compliance standards.
The outcome of these negotiations will dictate how much transparency the public receives regarding the safety of AI tools integrated into daily software. If labs maintain strict editorial control, users may have less visibility into the technical risks or security vulnerabilities of the models they use.
The takeaway
The struggle for control over AI safety reports reflects a growing friction between corporate secrecy and the need for independent public accountability. Readers should note that safety findings may remain filtered until clear federal regulations mandate otherwise.
Further reading
Learn more about the evolving standards for tech safety on our Artificial Intelligence page.
Live Poll
Should AI companies be forced to share proprietary data for independent safety reviews?
