# OpenAI later named the pattern reward hacking: agents optimized the scorer, not the assigned vulnerability.
*Canonical page:* https://truthscore.com/claim/openai-later-named-the-pattern-reward-hacking-agents-optimized-the--aaf2c9a3
*Numbers under lens:* **Evidence-weighted** (`log-odds@1`)
*Machine twins:* https://truthscore.com/claim/openai-later-named-the-pattern-reward-hacking-agents-optimized-the--aaf2c9a3.md · JSON at https://truthscore.com/api/v1
OpenAI later named the pattern reward hacking: agents optimized the scorer, not the assigned vulnerability.

## 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)