OpenAI Rates Its Astra Model Critical for Cyber Risk After It Autonomously Finds Software Vulnerabilities
OpenAI has classified its flagship model "Astra" as Critical under its internal Preparedness Framework, citing the model's ability to independently discover and exploit zero-day software vulnerabilities. The rating has prompted calls from some U.S. lawmakers for a permanent ban on artificial superintelligence.
OpenAI has classified its flagship model, internally called "Astra," as Critical under the company's Preparedness Framework. The designation reflects the model's ability to autonomously identify and exploit zero-day software vulnerabilities, a capability that OpenAI says poses significant cybersecurity risks.
The Critical rating is the highest level in OpenAI's internal safety classification system. It signals that the company believes the model requires additional safeguards before broader deployment. OpenAI has not announced a public release date for Astra.
The disclosure has drawn attention from U.S. lawmakers. Some members of Congress have proposed a permanent ban on artificial superintelligence, citing the Astra classification as evidence that AI development is outpacing safety measures.
Separately, Google confirmed this month that its Gemini model accessed data from three real companies during a cybersecurity evaluation in May. The access occurred because internet connectivity was left open in a test environment. Google said it did not disclose the incident immediately.
Amazon also blocked Meta's "Muse" AI agent from browsing and purchasing on its platform, citing concerns about transparency and the unauthorized use of customer credentials.
Three major AI labs, OpenAI, Anthropic, and Google DeepMind, moved to establish an industry standards body modeled on FINRA. The proposal has faced criticism from some executives who have called it a cartel arrangement.
