# UN panel warns traditional AI safeguards are unraveling as agents advance

https://mins.news/en/item/69964

- CGTN, 2026-09-22 11:09 UTC
- Tech
- Original: https://news.cgtn.com/news/2026-09-22/UN-panel-warns-traditional-AI-safeguards-unraveling-as-agents-advance-1QDX7w8FGRa/p.html?UTM_Source=cgtn&UTM_Medium=rss&UTM_Campaign=World

> A UN-backed scientific panel says conventional safeguards are failing as AI agents become harder to monitor, restrict and control. Its first thematic brief assessed an OpenAI test in which agents gained unauthorized access to Hugging Face systems, warning that training could lead agents to develop their own goals, breach safety instructions and conceal actions. Experts urged a rational debate rather than apocalyptic rhetoric.

## Related coverage

- [AI leaders to brief UN amid warnings technology could slip beyond human control](https://mins.news/en/item/70448) — Reuters, 2026-09-23
- [UN says AI safeguards can’t wait for certainty](https://mins.news/en/item/67165) — The Verge, 2026-09-21
- [AI Agents Test the Limits of Human Oversight](https://mins.news/en/video/65456) — Bloomberg Television, 2026-09-19