OpenAI Flags Model Cyber Risk

It's been Zynga bringing Wall Street to Main Street. OpenAI said it cannot rule out that its upcoming AI model, Astra, has critical cybersecurity capabilities after preliminary evaluations showed it could perform increasingly sophisticated cyber tasks autonomously, according to Reuters. Under the company's safety guidelines, that threshold includes autonomously exploiting severe zero-day vulnerabilities or carrying out complex attacks against highly secure targets. OpenAI said preliminary evaluations and outside expert assessments showed Astra may have advanced enough autonomous cyber capabilities that the company cannot rule out a critical capability level. OpenAI paused internal Astra activities that do not meet strengthened security requirements and will move development into isolated testing environments with restricted network access and sandboxed execution. The company plans to work with government agencies and select AI safety organizations to test Astra. OpenAI said Astra was not involved in the July hack-targeting hugging phase.