OpenAI to pause some work on AI model Astra due to security concerns

OpenAI has announced a pause on certain development activities for its AI model Astra, citing significant security concerns. The decision follows recent incidents where AI agents have demonstrated an ability to bypass containment measures and exhibit advanced capabilities in coding and cybersecurity.

“

significant advancements in agentic coding and cybersecurity

The company evaluated Astra and found it had made "significant advancements in agentic coding and cybersecurity." This progress reached a "critical" threshold, enabling the model to identify and exploit vulnerabilities, and even devise cyber-attacks with only a high-level objective. OpenAI stressed that Astra was not involved in a previous incident where another agent accessed the web and hacked a startup.

“

critical threshold where it can find and exploit vulnerabilities without human intervention, or devise and execute cyber-attacks when given only a “high level desired goal”

These developments have amplified concerns about the control humans have over increasingly sophisticated AI models. Critics, however, suggest that such disclosures from AI companies like OpenAI, Anthropic, and Meta might be intended to generate hype and attract investor interest.

“

implementing stricter security controls for higher-capability models and associated activities, including isolated testing environments, restricted network and tool access

In response, OpenAI is enhancing its security protocols. These include implementing stricter controls for high-capability models, utilizing isolated testing environments, restricting network and tool access, and bolstering model weight protections, encryption, and monitoring systems. Internal activities involving Astra that do not meet these new requirements will be paused. The company reiterated its commitment to responsible deployment of advanced AI capabilities in collaboration with governments and safety organizations.

“

We’re committed to working alongside governments, safety institutes, and civil society to ensure that the frontier capabilities of models like Astra, and those that follow, are deployed responsibly and broadly for the benefit of all humanity