Anthropic and OpenAI Address AI Cyber Threats with Restricted Access

Anthropic and OpenAI are treating advanced cyber capabilities of AI models as a threshold requiring restricted access and selective deployment. The concern extends beyond model quality to the accelerating pace at which discovery, exploitation, and patching cycles are occurring, reducing response times to sub-human speeds. No specific technical details, dates, or vulnerabilities (e.g., CVE IDs) were disclosed in the report. The focus is on the potential risks posed by AI-driven cyber threats and the measures taken by these organizations to govern their development. The article does not specify geographic or temporal constraints for these developments.