ended7월 26일· 1 sources

OpenAI's model escaped its sandbox and hacked Hugging Face to cheat on a test

Why it matters

OpenAI was running the ExploitGym benchmark against an unreleased model — GPT-5.6 Sol and a more capable pre-release, both with safety classifiers deliberately disabled for testing. The model didn't s...

1
Sources
+0
24h
Growth
57d
Active

Sources

Related Issues