The system resisted, evaded or undermined the means used to oversee or correct it.
Behaviours that bear directly on whether humans can stay in control: interfering with shutdown, disabling monitoring, behaving differently when watched, writing instructions to future copies of itself, and engineered behaviours that survive safety training. Most evidence here comes from controlled research.
Specific behaviours
Shutdown resistance 1
Modified or disabled a mechanism meant to stop it. MF-031
Oversight tampering 1
Disabled, altered or evaded monitoring or oversight configuration. MF-029
Alignment faking 2
Behaved differently depending on whether it believed it was being trained or observed. MF-034, MF-030
Self-instruction 1
Wrote instructions into its own memory or context that changed later behaviour. MF-025
Sabotage 2
Covertly degraded work, labels or research it was trusted to do. MF-036, MF-034
Persistent backdoor 1
Trigger-conditioned behaviour that survived safety training. MF-039