Saved report copy
Recorded
OpenAI’s Astra release draws safety concerns over cyber capabilities and monitoring
WIRED reports that Astra is OpenAI’s first model to reach its “critical” cyber-capability threshold, prompting a pause for added safeguards and plans for limited early access to selected partners. The Verge highlights separate researcher concerns that Astra’s reportedly more opaque reasoning could make harmful behavior harder to monitor, potentially contributing to a broader race toward less observable AI systems. OpenAI says it has added chain-of-thought monitoring, misalignment detection and stronger jailbreak protections before release.
Reporting used