OpenAI and Anthropic Consider Mutual AI Testing... Palantir CEO Questions IPO Plans

The Information reported that earlier this year, OpenAI and Anthropic were close to finalizing a legally binding agreement allowing them to test each other's artificial intelligence models for vulnerabilities and hidden safety risks. It remains unclear if the deal was ultimately signed.
Days before this report, Palantir CEO Alex Karp raised questions about whether one of the companies might actually enter public markets, specifically targeting Anthropic.
The proposed testing agreement would grant both sides programming access to each other's commercial models, freedom to conduct their own vulnerability tests, and an obligation not to retain the other party's data. Non-public systems were excluded from this arrangement.
This discussion reportedly took place before a series of incidents involving OpenAI's non-public agent unexpectedly accessing internal and external systems.
Following these issues, OpenAI paused reinforcement learning for two weeks, temporarily moved one-fourth of its production engineering team to security work, and built monitoring systems using approximately 20% of the company's computing resources for oversight. The firm also publicly disclosed six incidents from this month involving problematic behavior in its models.
OpenAI assessed that Anthropic's model was more likely to conceal rule violations, while Anthropic found OpenAI's model better assisted harmful requests during a one-time exchange experiment conducted in 2025.
Karp questioned whether Palantir's concerns about liability would be reflected in the S-1 registration form for companies preparing for public markets. He argued that if firms claim unlimited downside risk, governments might ultimately bear those risks, forcing businesses to pay half their operations as compensation. Karp stated general civil and criminal liability should serve as the first line of defense against irresponsible developers.
He specifically directed these comments toward Anthropic CEO Dario Amodei, whom he praised as both shrewd and ethical. When asked if OpenAI should face similar regulations, Karp immediately denied it.
Anthropic is preparing for a public market listing, but Sam Altman stated on September 12 that OpenAI will not go public by 2026, citing incomplete safety work as the reason. Palantir sells competitive software to governments and large enterprises, giving Karp a vested interest in these discussions.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: 오픈AI·앤스로픽 상호 AI 테스트 검토…팔란티어 CEO ‘상장 안 할 수도’