New York – Two of the nation’s biggest artificial‑intelligence developers, Anthropic and OpenAI, are pressing for greater control over what external evaluators may disclose. The companies argue that unrestricted publication could expose proprietary technology that underpins their trillion‑dollar valuation ambitions.
Companies want editorial safeguards
Anthropic CEO Dario Amodei told Reuters last month that outside reviewers should be free to publish key findings “without editorial control.” OpenAI later echoed the sentiment, saying assessors need “editorial independence.” Yet, sources close to the negotiations say both firms are now seeking strict editorial oversight to prevent accidental leaks of trade secrets.
According to the insiders, measures under discussion include lab staff monitoring embedded evaluation teams and prohibiting evaluators from working simultaneously at competing labs. Such safeguards aim to protect intellectual property while still allowing safety assessments.
Balancing safety and secrecy
The tension arises because the very data that AI labs deem proprietary often overlaps with information crucial for judging a model’s risk. Without clear standards, labs could dictate the scope of reviews, leaving evaluators dependent on the companies that built the systems they are assessing.
Experts warn that this power imbalance could lead to redactions that obscure genuine safety concerns. Rumman Chowdhury, former U.S. Science Envoy for AI under the previous administration and CEO of Humane Intelligence, noted that evaluators enjoy privileged access but “know it comes at a cost.”
Trump administration’s safety framework
President Donald Trump recently signed the White House Accord on Super Intelligence, a one‑page agreement that outlines four safety measures for advanced AI. The accord, which Trump described as “morally binding,” emphasizes independent evaluation, transparent methodology, and protection from retaliation. While the accord does not prescribe exact review procedures, it signals strong federal support for rigorous, independent oversight.
Anthropic has announced that Faculty, Accenture’s specialist AI business, will serve as its first embedded evaluator. Faculty CEO Marc Warner stressed that “evaluators are only useful if people can trust what they report, so independence and transparency are critical.” OpenAI, for its part, maintains that it will redact only confidential information and will allow evaluators to disclose when major redactions occur.
Critics and industry response
More than 200 researchers, including AI pioneer Geoffrey Hinton, have signed a public letter calling for independent control of findings, broad lab access, and safeguards against retaliation. The letter reflects a growing chorus of voices urging that safety reviews not be hampered by corporate secrecy.
Nevertheless, the negotiations illustrate a constructive dialogue between the federal government and industry leaders. By seeking a balance that protects trade secrets while upholding rigorous safety standards, the Trump administration aims to ensure that AI development proceeds responsibly and in line with American values.
What’s next?
The outcome of these talks will shape how the world scrutinizes powerful AI systems. As the industry moves toward potential initial public offerings, the pressure to safeguard both innovation and public safety will only intensify. Stakeholders on all sides agree that transparent, independent evaluation is essential for maintaining trust in a technology that could profoundly affect humanity.
Original reporting: Appleton, WI News Feed (HLL/CB) — read the source article.
