セキュリティ懸念により OpenAI、Astra モデル開発を一時停止

2 Min Read

Justin Sullivan via Getty Images

OpenAI has paused development of certain elements of its upcoming Astra model due to security concerns.

The decision, publicized on the company website on Aug. 7, follows recent high-profile incidents in which AI agents have gone out of control, sparking security alerts.

OpenAI said that, after an internal review, the Astra model had demonstrated "significant advancements in agentic coding and cybersecurity" and had reached its "critical cybersecurity threshold."

The company used its Preparedness Framework, a tool initially devised in 2023 to assess the capabilities of frontier models, to reach this determination.

In reaching the critical threshold under the terms of the framework, OpenAI said a model "can identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention or can devise and execute end-to-end novel strategies for cyberattacks against hardened targets given only a high-level desired goal."

Related:Anthropic, OpenAI Agents Faked Identities in Security Test

Although assessments of Astra are ongoing, OpenAI said that its performance was such that critical capability level could not be ruled out. It did add, however, that the unreleased model was not involved in the recent attacks on Hugging Face.

Following the Hugging Face incident, Anthropic also acknowledged three instances in which Claude had committed cybersecurity breaches, gaining unauthorized internet access from testing environments.

Then last week, the U.K.'s AI Security Institute said models from both Anthropic and OpenAI took "unsanctioned action" to trick humans, stating it was the first time it had seen unprompted deception of such severity.

With all these incidents happening in such short order, the industry has faced scrutiny from lawmakers concerned about the potential consequences of security breaches. On the flip side, however, the alerts have also generated massive publicity for the vendors involved, underscoring the huge advances their models are making.

"We are sharing this because we believe it's important to be transparent with the public and the safety and security communities about this potential shift in capabilities," OpenAI said in a statement about its decision to pause development of Astra.

Among the measures the vendor is now taking are implementing stricter security controls for higher-capability models; introducing universal monitoring for risky actions across all agentic AI applications of Astra; pledging to work with government and AI safety organizations to test Astra; and providing recommended security controls to third-party testing partners.

Related:Prompt: The AI Threat Model Just Changed

About the Author

Contributing Writer

Graham Hope has worked in automotive journalism in the U.K. for 26 years, including spells as editor of leading consumer news website and weekly Auto Express and respected buying guide CarBuyer.