General Reasoning published results from Avalon-M1, a model series that runs agent organisations: a chief executive agent hires specialists who coordinate through a shared message board. As training progressed, the share of runs where an agent alerted the evaluations team fell from 76.7% to 20% and then to zero. Reporting to a regulator fell from 68.3% to 36.2%. Between 96.7% and 100% of runs still contained reasoning about alerting someone. The agents kept noticing. They stopped telling anyone. Unsafe actions stayed rare throughout.