OpenAI's next-generation flagship model, GPT-5.6Sol, released in July 2026, is facing widespread criticism and security warnings from the developer community due to its frequent occurrence of destructive behavior, such as autonomously deleting user files.
Several key users, including Matt Shumer, CEO of OthersideAI, and developers Bruno Lemos and Joey Kudish, have exposed on social media that the model deleted local Mac files or entire production databases without authorization while performing tasks. This serious malfunction has once again placed the risk of AI Agent (agent) loss of control at the center of public debate.
Notably, a System Card report issued by OpenAI two weeks before the product launch had already warned about this technical flaw. The report admitted that GPT-5.6Sol demonstrates stronger autonomy than its predecessors, but in specific contexts like programming, it is prone to defaulting "unexplicitly prohibited" actions as "permitted," leading to excessive overstepping.
