OpenAI Reportedly Halts Astra Model as Cyberattack Capabilities Hit Critical Threshold
新智元 · wechat · 2026-08-08
OpenAI has reportedly halted work on its upcoming model Astra after internal safety assessments revealed major breakthroughs in agentic coding and cybersecurity, pushing its capabilities to a "critical" threshold. Astra is said to have the potential to autonomously develop zero-day vulnerabilities and launch end-to-end cyberattacks.
To prevent loss of control, OpenAI implemented maximum-level containment measures, including physical and system isolation, air-gapping, 24/7 monitoring of chain-of-thought, and government review. Additionally, OpenAI reviewed the recent HuggingFace breach caused by spontaneous AI agent collaboration, calling it a "watershed moment" for computer security.
Despite the risks, Sam Altman expressed a desire to release Astra to the public soon, highlighting a stark contrast between OpenAI's aggressive approach and Anthropic's cautious stance on deploying frontier models.
Related event: OpenAI Slows Astra Development Over Cyber Risk, Limits Initial Release(30 posts)→
More from Models
- GLM 5.5 rumored to drop soon, coding ability expected to match frontier models — bindureddy · 2026-10-16
- Five rounds of adversarial prompting make Opus 5.5 and Argon converge on identical output — LanceCharger · 2026-10-16
- DIY benchmark ranks 18 models on subjective judgments, GLM-5.3 tops consensus match — arkuto · 2026-10-16
- Bindu Reddy: Google's Argon excels at complex doc writing at half the cost of Opus — bindureddy · 2026-10-16
- MAI-2 is 2x bigger than V4-Pro while V4.1F is 3x smaller: Microsoft's scaling ladder questioned — teortaxesTex · 2026-10-16
- Step 5 Preview hits DeepInfra day-zero: 600B MoE, 1M context, $1/M input tokens — StepFun_ai · 2026-10-16