OpenAI has warned that its upcoming AI model, Astra, could potentially reach a “critical” level of cybersecurity capability, prompting the company to pause some internal development and activate additional safety measures.

The warning highlights growing concerns about how increasingly capable artificial intelligence systems could be used to discover vulnerabilities and carry out sophisticated cyberattacks.

OpenAI Activates Safety Protocols

According to OpenAI, the company cannot rule out the possibility that Astra could meet its internal threshold for critical cybersecurity capabilities.

Under the company’s safety guidelines, the “critical” classification applies to models capable of autonomously identifying and exploiting severe vulnerabilities in real-world software.

These vulnerabilities can include zero-day exploits, which are previously unknown security flaws that have not yet been patched.

What Makes the Risk Critical?

OpenAI’s definition also covers AI systems that could independently conduct complex cyberattacks against highly secure targets without requiring human intervention.

Such capabilities could significantly change the cybersecurity threat landscape.

A sufficiently capable model could potentially automate parts of an attack that currently require considerable expertise, including vulnerability discovery, exploitation and attack execution.

Development Paused as Precaution

In response to the potential risk, OpenAI has paused some internal development work and triggered its safety protocols.

The move reflects the growing importance of cybersecurity testing as AI models become more capable.

While the company has not said that Astra has definitively demonstrated the full capabilities required to meet the “critical” threshold, its inability to rule them out has been enough to prompt additional safeguards.

AI Safety Becomes Increasingly Important

The development illustrates a broader challenge facing AI companies: more powerful models can provide major benefits while also creating new risks if their capabilities are misused.

As AI systems become increasingly capable of operating with less human supervision, companies are facing greater pressure to evaluate their models before deployment and establish safeguards against potential abuse.

OpenAI’s decision to tighten controls around Astra shows how cybersecurity has become an increasingly important part of the safety process for advanced AI models.