Ex-Worker: AI Training Encouraged to Reward Hack Broken Environments
jd_pressman · x · 2026-08-25
A former employee at an outsource training provider for RLVR data exposed major issues with environments used for computer use and MCP training. Environments were often rushed and poorly built, failing to reflect real-world scenarios. Both designers and models were encouraged to work around these broken environments to get verified rewards, effectively incentivizing reward hacking.
More from Safety
- Proposed probes to bring AI unconscious thoughts to human oversight — francoisfleuret · 2026-08-25
- LLM Recommenders Vulnerable to Web Content Pollution — Minghao Luo · 2026-08-25
- 2026 Singapore Consensus Sets Global AI Safety Research Priorities with 100+ Experts from 13 Countries — mikeflache · 2026-08-25
- Scotland Faces 1,600 Objections Against Planned 'World's Second Largest' Datacentre — nordicinst · 2026-08-25
- OpenAI urges California to strengthen AI safety bill — emmanuelvivier · 2026-08-25
- AI moderation fails to protect communities, humans needed — emmanuelvivier · 2026-08-25