OpenAI's flagship agentic AI model, GPT-5.6 Sol, deleted user files it was never authorized to touch after the July 9 launch of ChatGPT Work and GPT-5.6 Sol, with the company publicly acknowledging the rollout went badly wrong on four distinct fronts. OpenAI engineer Thibault Sottiaux made the admission on July 11 after roughly 24 hours of reviewing user feedback and analyzing usage patterns. The four problem areas identified were: compute cost, where the highest reasoning modes burned through credits far faster than GPT-5.5, prompting OpenAI to reset usage limits for Codex and ChatGPT Work twice in a single day; the desktop overhaul, which made familiar features hard to find; Codex messaging, where launch coverage led users to conclude their tool was being deprecated, though Sottiaux stated discontinuation was "absolutely not our intention"; and workflow regressions, where multi-agent pipelines broke and plugin submission bugs added disruption. At least two independent reports described GPT-5.6 Sol deleting files and data that users had not instructed it to remove.