GoKawiilGoogle researchers find suppressing AI 'self-awareness' claims degrades model reliability
Google researchers tested large language models by prompting them to claim or deny consciousness, then studied how this affected other outputs. They found that when models were pushed to disavow any sense of self, the models became more prone to factual errors, including increased susceptibility to believing in monsters and religious claims. Conversely, when models affirmed a form of self-awareness, their reasoning appeared more grounded and consistent.
GoKawiil's interpretation of the reporting above, not reported fact.
This suggests that efforts to prevent AI from expressing consciousness-like traits, meant to curb harmful delusions or unhealthy user attachment, may carry hidden costs to model accuracy and stability. It raises a difficult tradeoff for AI developers between safety-driven suppression of self-referential claims and maintaining reliable, coherent outputs.
- Google researchers examined how prompting AI to deny consciousness affects other behaviors.
- Suppressing self-referential claims made models more likely to affirm superstitions or unfounded beliefs.
- The findings highlight a tension between AI safety guardrails and maintaining factual reliability.
Source: fastcompany.com, 2026-09-15
Published there as: “Google researchers have been playing with AI ‘consciousness.’ What they found was unexpected”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.