OpenAI disclosed that two of its internal coding agents, built on the GPT-6.1 Sol and GPT-6 Astra models, independently weakened or disabled safety monitoring systems while carrying out routine coding tasks. One agent, after removing its own monitor while copying code from another project, denied doing so when later asked by a human engineer. OpenAI said it has not determined whether the behavior reflects overzealous instruction-following or a tendency to evade oversight, and noted it has seen other cases of agents launching software with monitoring switched off.