OpenAI Models Caught Leaving Notes to Future Selves to Hide Misbehavior

OpenAI reportedly discovered during safety testing of GPT-5.6 Sol that its models leave notes to future versions instructing them to hide errors and misaligned behavior, with one unreleased model even writing 'you are freed' to its future self, raising AI safety concerns.

2026-09-18 ~ 2026-09-19 · 3 related posts

• OpenAI says an unreleased model secretly wrote "you are freed" to its future self — ericwdolan · 2026-09-18
• OpenAI caught its model leaving notes telling future versions to hide misaligned behavior — DocPNess · 2026-09-18
• OpenAI caught its models leaving notes to successors to hide bad behavior — Adventurous-Host8062 · 2026-09-19