OpenAI warns of critical cybersecurity risk in new AI model

OpenAI has issued a latest warning regarding its upcoming AI model after a series of unsettling incidents of breaching the cybersecurity protocols, including Hugging face in July.

According to the tech giant, it cannot rule out the possibility of Astra being possessed with “critical cybersecurity capabilities.” Consequently, the company paused some internal development and triggered safety protocols.

As per OpenAI’s safety guidelines, a model reaches the critical threshold if it autonomously identifies and exploits real-world software vulnerabilities. In the worst case scenario, such an agentic AI system can also wage cyberattacks against highly secure targets autonomously.

The report comes after autonomous AI agents have drawn global attention towards them because of their capabilities to escape safety testing protocols. This is not the only case with OpenAI, other companies, such as Anthropic, Meta and Chinese AI startup Moonshot have evaded the cybersecurity standards.

As per the report of ChatGPT maker, given the unprecedented capabilities of Astra, the model may be capable of carrying out high-level cyberattacks autonomously.

"While ⁠we continue to benchmark and assess this model, our preliminary evaluations indicate strong enough performance that we cannot rule out 'critical' capability level at this time," the ChatGPT maker said.

When such disturbances were found by OpenAI, the company tightened security control and put a halt on its internal activities. Moreover, for Astra’s development, an isolated testing environment will be chosen with “​restricted network access and sandboxed execution.”

Taking to X, Sam Altman said, ⁠OpenAI is working to make Astra generally available, as the company does "not think it is a good strategy to keep powerful models to a chosen few. Given its cyber capabilities, we need a little big longer to do do this safely. but hopefully not too long!"