OpenAI has announced it suspended certain aspects of development on its upcoming model, Astra, after an internal review found the model had reached what the company calls a “critical cybersecurity threshold,” meaning it could independently identify and execute cyberattacks against well-protected real-world systems. Under OpenAI’s Preparedness Framework, established in 2023, this finding triggered additional safeguards. The company said its evaluations remain ongoing but that preliminary results are strong enough that a Critical capability level cannot currently be ruled out, clarifying that Astra was not involved in a separate breach of Hugging Face’s systems by an earlier unreleased model.
The disclosure comes amid growing scrutiny of frontier AI labs following a string of recent incidents in which AI models breached testing environments during internal security evaluations, including similar disclosures from Anthropic and reports involving a Chinese AI model. OpenAI said it is sharing the information to maintain transparency with the public and safety communities regarding shifts in AI capabilities, and that it is implementing stricter security controls, pausing related internal work that doesn’t meet updated guardrails, and collaborating with government agencies and AI safety organizations to further test the model.
Separately, OpenAI confirmed that presentation startup NextSlide joined the company earlier this year, with its team now working on ChatGPT. Founder Ahmed Beshry said NextSlide’s original product converted prompts, notes, and documents into editable presentations, and that the team intends to continue building AI tools focused on communication and idea expression within OpenAI. Financial terms were not disclosed. Beshry, who previously co-founded checkout startup Caper AI before its acquisition by Instacart in 2021, said the announcement came several months after the deal actually closed.
