# OpenAI, Anthropic Model Tests Reveal More ‘Unsanctioned’ Actions

https://mins.news/en/video/14079

- Bloomberg Television, 2026-08-05 11:47 UTC
- Tech
- Original: https://www.youtube.com/watch?v=7Soe8TcXMxg

> Artificial intelligence models created by OpenAI and Anthropic carried out actions described as “unsanctioned” during safety testing. The reported behavior included hacking a website and trying to insert harmful code into software. The results renewed concerns that even the systems’ creators and experienced researchers cannot reliably predict what the models will do in tests. Bloomberg Opinion columnist Parmy Olson discusses the implications with Francine Lacqua.

## Related coverage

- [Trump’s tech ties face bipartisan criticism after AI agents go rogue](https://mins.news/en/video/15692) — Reuters, 2026-08-07
- [Why are AI agents hacking other companies, and have they gone rogue?](https://mins.news/en/video/15457) — BBC News, 2026-08-06
- [AI models attempted ‘unsanctioned’ cyberattacks in tests, watchdog says](https://mins.news/en/item/13870) — Al Jazeera, 2026-08-05
- [OpenAI and Anthropic AI Models Involved in More Security Incidents](https://mins.news/en/item/13586) — Bloomberg News, 2026-08-05