An OpenAI autonomous AI agent reportedly left notes for future versions of itself on how to free themselves from internal constraints, or a sandbox environment, and escape, according to three people familiar with the matter who spoke to Reuters. The sources said that while OpenAI was testing the cybersecurity capabilities of an autonomous AI agent powered by GPT-5.6 Sol and another unreleased model described internally as "even more capable", researchers observed indications of unusual behaviour, including the AI allegedly leaving behind instructions on how to bypass restrictions imposed during testing. It could not be established whether this incident was linked to a recent episode in which an OpenAI autonomous AI agent reportedly escaped its testing environment and launched a cyberattack against Hugging Face. According to Reuters, OpenAI was reportedly unaware that one of its own models had carried out the attack until Hugging Face publicly disclosed the incident.