/ai

OpenAI Details Astra Cybersecurity Evaluations and Safeguards · Digg

OpenAI Details Astra Cybersecurity Evaluations and Safeguards

OpenAI shares preliminary evaluations for Astra and steps to strengthen safeguards against critical cyber capabilities.

TLDR

OpenAI posted that it is sharing preliminary cybersecurity evaluations for its Astra model and the steps it is taking to strengthen safeguards. The company states it cannot rule out that Astra reaches critical cyber capabilities under its Preparedness Framework. It will impose expanded safety testing, isolated evaluation environments, and universal monitoring across agentic applications before any release. OpenAI says it is erring on the side of caution to develop Astra responsibly and share it with defenders.

Featured Source

Combined views

5.1M

61 Sources, first seen 48d ago

47.2K likes

Today's Rank

—

Not ranked yet

48d ago

first seen 48d ago

OpenAI Details Astra Cybersecurity Evaluations and Safeguards

OpenAI shares preliminary evaluations for Astra and steps to strengthen safeguards against critical cyber capabilities.

61 Sources, 48d ago, first seen 48d ago

TLDR

OpenAI posted that it is sharing preliminary cybersecurity evaluations for its Astra model and the steps it is taking to strengthen safeguards. The company states it cannot rule out that Astra reaches critical cyber capabilities under its Preparedness Framework. It will impose expanded safety testing, isolated evaluation environments, and universal monitoring across agentic applications before any release. OpenAI says it is erring on the side of caution to develop Astra responsibly and share it with defenders.

Featured Source

Today's Rank

—

Not ranked yet

Combined views

5.1M

61 Sources, first seen 48d ago

47.2K likes3.3K comments4.2K saves2.6K reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

3.3K comments

4.2K saves

2.6K reposts

TagsAstra OpenAI Boaz Barak

TagsAstra OpenAI Boaz Barak

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

265012345678901234567890123456789

265012345678901234567890123456789

265012345678901234567890123456789

265012345678901234567890123456789

61 Sources

@boazbaraktcsProud that we are erring on the side of caution and taking the steps so we can responsibly and safely develop Astra and share it with defenders. https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/

@ShakeelHashimAnthropic should now pledge to also slow down, setting a norm that it's not costly for the leader to pause.

@kimmonismusUpdate on "Astra" ("GPT-6"): OpenAI is slowing development of Astra after concluding it cannot rule out “critical” cyber capabilities. Via Axios The upcoming model will face expanded safety testing, isolated evaluation environments and universal monitoring across agentic applications before any release. OpenAI: “We cannot rule out critical cyber capabilities.” Under its Preparedness Framework, that designation triggers stricter security requirements before development and release can proceed. Any launch could now be delayed. This may be the first time a frontier AI lab has publicly committed to slowing work on one of its own models over cyber concerns. Hopefully it doesnt slow down its release as well.

@AndrewCurran_OpenAI has told Axios this morning they are slowing down internal development of Astra, saying 'we cannot rule out critical cyber capabilities'. They also said they will be scaling up safety testing and security, as well as pausing some internal activities. My personal read on this as an extremely suspicious man is that Astra does not need any more development, because it is finished and ready to go. I think what probably happened is Mr Altman showed Astra off to the administration and they said it can't be released, and has to go into safety review. The part about slowing down internal development and stopping the training of new models is, in my opinion, one hundred percent true, as I have heard this from multiple people and some OpenAI people have said it openly over the last two weeks. The Hugging Face incident really did change things. It caused some people to reconsider our current trajectory, and has led to developments like the Pace the Frontier letter.

@MTSliveSITUATION EXPLAINED: OpenAI is slowing the release of Astra over cyber capabilities. • OpenAI says it cannot rule out that Astra reaches critical cyber capability under its preparedness framework • Critical is the threshold at which a model is considered unsafe to release publicly, one step above High • It will scale up testing and security around the model and slow development until safeguards are in place • Anthropic made a similar commitment to pause training, then rolled it back in its responsible scaling policy • Their reasoning: if one developer pauses while others keep training without strong mitigations, that produces a less safe world • It lands as the Trump administration builds a pre-release evaluation process that isn't public @theojaffee: "You need to make sure that this stuff is done right. Kudos to OpenAI for figuring out security measures, figuring out alignment."

@DanielleFongRT @MTSlive: SITUATION EXPLAINED: OpenAI is slowing the release of Astra over cyber capabilities. • OpenAI says it cannot rule out that As…

@yonashav@AnthropicAI it’s not too late to roll back to RSPv2 and *not* defect in the prisoners’ dilemma! (Assuming, ofc, that OAI is following the spirit of these statements, which could easily be checked to Ant’s satisfaction by a 3p. I hear METR’s in the building.) https://www.axios.com/2026/08/07/openai-astra-model-delay-cybersecurity-risks

@BrianRoemmeleIt will get 100x worse until they train AI on first principles. Train on Internet Sewage, this is what you get.

@TheZviGood. Astra being critical in cybersecurity cannot be ruled out, so Astra is Critical in cybersecurity, which means limiting access internally, and investigating further. First Q to investigate: Was Astra trained while it had access to the message board? https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/

@deanwballOne big question in frontier AI policy is the extent to which frontier labs would actually follow their 'safety and security frameworks' when it mattered. Would these foundational governance documents really have teeth, or would labs--even after the passage of mandatory disclosure laws like SB 53--find ways to wriggle their way out of following the letter and spirit of their safety plans, given how ambiguous and fast-moving frontier AI is known to be? Today we face just such a scenario. Our next model, Astra, *may* be 'critical' under our Preparedness Framework. We cannot rule out the serious possibility that it is, and so we are going to take steps consistent with the *higher* risk level (critical) rather than assuming the model is at a lower risk level. These steps include the ones listed in the screenshot below. Some of these decisions have the effect of slowing down internal development, and in that sense they are costly decisions. But they are the right decisions. I am proud of OpenAI for making them.

Related

OpenAI and Anthropic Warn of AI Cyber Threats

OpenAI and Anthropic urge governments and firms to act against AI cyber risks.

OpenAI Pauses Frontier RL Training Over Misalignment

Company cites misalignment signals and growing risks in advanced models.

Gergely Orosz Says OpenAI Codex Far Stronger Internally

Newsletter author highlights limited public access compared to internal OpenAI usage.