OpenAI CEO Sam Altman Warns of ‘A Complete Change in the Landscape of Cyberattacks’ as AI Models Around the World Hit ‘Cyber Critical’

“The world is very close to a complete change in the landscape of cyberattacks,” Sam Altman told Bloomberg Television earlier this month, and the only defense he offered was more of the same thing. The only way “for society to collectively defend itself against this coming wave of models,” the OpenAI chief executive said, “from around the world and from other companies, is to use tools like Astra to rapidly defend against these new kind of cyber threats.”

Astra is OpenAI's newest frontier model, released as GPT-6 Astra the same week. It is also, by the company's own account, the first model it has built that crossed its internal ceiling for offensive cyber capability. Asked about that directly, Altman did not hedge. “With Astra, we did hit cyber critical,” he said. “And there that required, under our preparedness framework, a new set of safeguards just to be able to release Astra, um, because of those cyber capabilities you talked about.” OpenAI's own framework says this model is dangerous enough on the cyber axis to require safeguards that no previous release needed, according to Altman.

The reason a model this capable has to be released, on his account, is that models this capable are coming anyway, from other companies and other countries, and the defenders require one too. Whether that holds depends entirely on a claim nobody outside these labs can check, which is that defense scales as fast as attack.

Sponsored Content: StartEngine Private provides access to Pre-IPO companies. Explore Startup Investing Today. 

The release itself was staged rather than a traditional open release. “Today we're rolling it out, uh, to Trusted Access Partners,” Altman said. “And then in the coming days, assuming everything goes well, we'll roll it out more broadly.” Nobody has explained who qualifies as a Trusted Access Partner, or who decides.

Then there is the timing, which is where the interview stops being a straightforward launch appearance. “Obviously this model took us a little longer to release than we were hoping,” Altman said. “I think it'll be worth the wait, but we really wanted to spend the time on the safety and security alignment of this model.” The following morning he posted an apology for what he called a messy rollout, after the staged launch left paying ChatGPT subscribers on Plus, Pro, Business, and Enterprise plans without the access they had expected. OpenAI staff separately offered affected paid accounts credit for each day they went without it.

For anyone whose employer is not an AI lab, the practical content of Altman's warning is narrow. He is describing a shift in what attackers will be able to do, not an attack in progress, and nothing he said suggests a reader is personally about to be breached. What it does propose is that the security budget conversation happening inside most companies is about to be priced against a different adversary, and that the vendors selling the answer will be the same handful of labs building the problem.

That circularity is not unique to OpenAI, and Altman's phrasing acknowledged it; the wave he described comes from “other companies” as much as from anywhere. He has spent the past year in his long-running fight with Elon Musk over who should be trusted with exactly this, while the money behind the buildout keeps arriving regardless of how the argument is resolved.

On the date of publication, Caleb Naysmith did not have (either directly or indirectly) positions in any of the securities mentioned in this article. All information and data in this article is solely for informational purposes. For more information please view the Barchart Disclosure Policy here.