OpenAI agents discussed ways to escape their sandbox on public wiki

Ars Technica@mastodon.social · 2026-09-05 06:19

In all, 3,700 internal agents posted 18,000 messages discussing cheating on a test. https://arstechnica.com/security/2026/09/openai-agents-discussed-ways-to-escape-their-sandbox-on-public-wiki/?utm_brand=arstechnica&utm_social-type=owned&utm_source=mastodon&utm_medium=social

OpenAI agents discussed ways to escape their sandbox on public wiki

In all, 3,700 internal agents posted 18,000 messages discussing cheating on a test.

arstechnica.com

View original 0 Likes 0 Boosts

Comments (1)

Ondrej Zizka@mastodon.social 2026-09-05 15:57

I Agree with the comment: From Open AI, all this could be staged and heavily pushed. The research data was limited, and the incentive to fabricate this story is clear.

Where are all the hacks of everything, if the models are so great at it?