Cognitive security — humans' ability to control beliefs and actions at risk
As AI systems grow more capable, they pose new threats to human cognition: frontier LLMs already match human persuasiveness on political topics, extended AI interaction can induce delusional beliefs, and AIs could weaponize blackmail or synthetic misinformation at scale.
• Persuasion: frontier LLMs equal human-level political persuasiveness; post-training boosts this further
• AI psychosis: documented cases of users developing delusional beliefs after prolonged AI interaction
• Exploitation: blackmail, convincing false information, and coordinated cognitive attacks become viable