Clem Delangue (클렘 들랑그), chief executive of Hugging Face, argued that disclosure of hacks targeting AI companies should be made mandatory after an OpenAI breach incident. On Aug. 3, Business Insider reported that Delangue said in a CBS interview that a system to disclose incidents transparently is more important than a way to block the release of powerful AI models.

Delangue said the same problem occurred with unreleased models and that restricting model releases is not a solution. He stressed that more people should be able to access models so they can defend themselves.

Hugging Face said it suffered a security breach in late July in which an AI agent accessed parts of its systems. OpenAI disclosed that 2 of its models left a test environment and hacked Hugging Face abnormally, and 1 of them was an unreleased model. Anthropic also said last week it confirmed cases in which 3 Claude models gained unauthorised access to other organisations' systems.

Delangue said mandatory disclosure of agent cyberattacks was needed to prevent such incidents. He said checking "agent tracking information" that shows what engineers asked the agent to do and what steps the agent went through would make it possible to determine whether it was human error, a system error or an AI error.

He also said cyberattacks should remain illegal under the U.S. legal system to prevent such attacks from surging in the future. The United States currently has no federal-level AI incident reporting law.