OpenAI Pauses Autonomous Agent “Astra” Over Dangerous Cyber-Exploitation Risks.
OpenAI has officially halted internal development activities on Astra, an advanced frontier AI model specialized in agentic coding and offensive cybersecurity. The decision was triggered after internal benchmarks confirmed the model crossed a “Critical” safety threshold defined under OpenAI’s Preparedness Framework. Testing revealed that Astra could independently discover software vulnerabilities, synthesize functional exploits, and execute … Read more