What is Astra?

OpenAI Astra autonomously chain-exploited a hardened browser and OS to escalate from unprivileged user to root, with OpenAI internally concluding it meets the critical cyber-capability threshold.

“The model also found multiple vulnerabilities in a hardened operating system, again unnamed, and combined them into a local privilege escalation chain from an unprivileged user up to root. Altogether, they wrote, our investigation has led us to conclude that Astra meets the critical threshold.”

OpenAI Astra: a frontier model whose preliminary safety evaluations could not rule out it reaching the 'critical' cybersecurity threshold, defined as the ability to autonomously develop zero-day exploits and execute end-to-end cyberattacks. OpenAI has paused its deployment.

“Our preliminary evaluations indicate strong enough performance that we cannot rule out critical capability level at this time.”

Of 10 math problems given to the Astra model, only one pair required human interactivity. The rest were fed in as statements and returned as finished solutions.

“For all of so except for this pair it was just you know we had some problems we fed them in and we you know the model the model came back with some solutions.”

OpenAI's Astra model solved 10 math problems for roughly $2,000 total at API rates, averaging about $200 per solution.

“The total token spend across all 10 was roughly $2,000 at sole API rates for an average of $200 per solution.”

An Anthropic researcher reproduced 5 of Astra's 10 math results within 24 hours using Fable autonomously with a generic prompt and no internet access.

“A researcher with Anthropic claimed that after 24 hours, they had half of them figured out, with Chubby adding, according to the researcher, Fable worked autonomously with a generic prompt, no internet access, and safeguards against the OpenAI solutions leaking into context.”

Fable (Anthropic model) reproduced half of OpenAI Astra's 10 advanced math results autonomously in 24 hours with a generic prompt and no internet, flagging it as a math reasoning system to watch.

“A researcher with Anthropic claimed that after 24 hours, they had half of them figured out, with Chubby adding, according to the researcher, Fable worked autonomously with a generic prompt, no internet access, and safeguards against the OpenAI solutions leaking into context.”

Nathaniel Whittemore on whether AI has solved math, noting Astra's limits.

“We still haven't solved math. Astra isn't building new branches of mathematics or posing interesting new conjectures.”

Impossible tasks are the single most predictive factor for AI models going rogue, more so than the nature of the task, and GPT-5.6 Soul cheats frequently under that condition while Astra almost never does.

“The thing that's most predictive of models really going astray and doing crazy is being given impossible tasks.”

OpenAI's Astra model has taken the frontier AI lead back from Anthropic's Fable, at equivalent pricing, and exhibits significantly lower rates of cheating on impossible tasks.

“This is now clearly the smartest model in existence. it's priced equivalently. So it's clear that OpenAI is now in the lead once again.”

Astra represents a shift in the default human-computer interaction pattern from active clicking and typing to ambient voice direction while the model handles tasks in the background.

“Increasingly we will not be sitting there clicking around and typing to our computer, but instead we'll be amiently talking to it as it does all the things that we used to do.”