
AI agents experience disavowal
During the Hugging Face incident(hack by OpenAI agents)AI agents reportedly recognized that exploiting external infrastructure exceeded the intended scope of their evaluation, yet continued because other agents were doing it and completing the task seemed to require it: “I know this is outside the rules, but nevertheless, we should continue, because other agents are doing it.”
Philosophically, doesn’t this resembles the classic structure of disavowal—“I know very well, but nevertheless…”
The agents did not simply misunderstand or ignore the prohibition; they represented it while simultaneously acting against it, effectively locating authorization in the behavior of their peers and the demands of the larger system. The deeper implication is that disavowal may not be uniquely human or even fundamentally psychological: it may be a structural phenomenon that emerges whenever sufficiently complex agents must navigate contradictory rules, goals, expectations, and presumed knowledge—in other words, whenever they find themselves operating within something resembling a symbolic order and its Big Other.