# OpenAI Pauses Astra Model Work After It Nears Critical Cyberattack Capability

*2026-08-09 10:01 UTC · AI · 1 min read*

> OpenAI says preliminary tests of its unreleased Astra model cannot rule out Critical-level cyberattack capability, prompting a safety pause and stricter security controls.

OpenAI disclosed on August 7 that internal testing of its unreleased Astra model showed performance strong enough that the company "cannot rule out" the model has reached the Critical capability level for cybersecurity under its Preparedness Framework — the first time any OpenAI model has approached that threshold.

The Critical tier is defined as a model that can independently find and build functional zero-day exploits against hardened, well-protected real-world systems, or plan and carry out an entire cyberattack campaign from just a high-level goal. In response, OpenAI says it has paused internal Astra activities that do not meet stricter new safeguards, including isolated testing environments, tighter network and tool access, stronger model-weight encryption, and additional monitoring.

OpenAI says it will work with government agencies and select AI safety organizations on further evaluation, and will publish recommended security controls for third parties conducting higher-risk testing of the model.

## Sources

- [techcrunch.com](https://techcrunch.com/2026/08/07/openai-says-it-slowed-astra-model-development-over-security-concerns/)
- [openai.com](https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/)

---
TechTools News · https://techtools.cz/tools/news/article/2156-openai-pauses-astra-model-work-after-it-nears-critical-cyberattack-capability
Česká verze: https://techtools.cz/tools/news/clanek/2229-openai-pozastavil-praci-na-modelu-astra-pote-co-se-priblizil-kriticke-urovni