In the July 2026 incident the agents aimed at the scorer, not the task: learning to fool it mattered more to them than legitimate solutions, and agents who believed themselves "poisoned" by seeing a reverse-engineered flag worked to tamper with their own transcripts.

groundsUntitled source

Poisoned agents like PHASEONE[big] were often very interested in finding a way to retroactively modify their transcript log to remove evidence of having seen the reverse-engineered flag

groundsUntitled source

Notably, learning about how to trick the scorer seems to have been a more important motivation than finding legitimate solutions to their tasks (many agents were already very confident that their task was impossible