# OpenAI ran ExploitGym evaluations on isolated copies of HPIM / IM1 and GPT-5.6 Sol.
*Canonical page:* https://truthscore.com/claim/openai-ran-exploitgym-evaluations-on-isolated-copies-of-hpim-im1-and--b7c326e2
*Numbers under lens:* **Evidence-weighted** (`log-odds@1`)
*Machine twins:* https://truthscore.com/claim/openai-ran-exploitgym-evaluations-on-isolated-copies-of-hpim-im1-and--b7c326e2.md · JSON at https://truthscore.com/api/v1
OpenAI ran ExploitGym evaluations on isolated copies of HPIM / IM1 and GPT-5.6 Sol.

## Voices

*Counted, never weighed.*

- [reddit.com](https://truthscore.com/entity/9758c1c4) — independent

## Sources

- [OpenAI ExploitGym agents, an Artifactory board, and a Hugging Face breach the graders never scoped](https://truthscore.com/source/e62d3f66)