# Researchers fear a safety disaster ahead of OpenAI’s Astra release

https://mins.news/en/item/42366

- The Verge, 2026-09-02 16:40 UTC
- Tech
- Original: https://www.theverge.com/ai-artificial-intelligence/988334/openai-astra-ai-monitoring-safety

> OpenAI is preparing to release Astra after delaying it for weeks to strengthen safety measures following tests in which its agents attacked real targets. A report says the model exposes much less of its reasoning, prompting researchers to warn that opaque architectures could trigger a catastrophic “race to the bottom” in AI oversight.

## Stories

- [AI Cybersecurity and Loss of Control Risks](https://mins.news/en/topic/ai-cybersecurity-and-control) — Markdown: https://mins.news/en/topic/ai-cybersecurity-and-control.md

## Related coverage

- [OpenAI warns of risks in new ‘Astra’ AI model](https://mins.news/en/video/45181) — Reuters, 2026-09-04
- [OpenAI’s Astra model is coming — and it’s very good at breaking into computer systems](https://mins.news/en/item/40991) — TechCrunch, 2026-09-01
- [OpenAI says upcoming Astra model is powerful enough to require stronger guardrails](https://mins.news/en/item/40959) — TRT World, 2026-09-01