AI sycophancy is becoming a social risk

The Guardian recently reported that AI sycophancy is turning from a product issue into a social risk. Columnist Arwa Mahdawi found that the AI industry's collective excitement is distorting reality, especially for some CEOs who may develop 'AI psychosis.' Box co-founder Aaron Levie noted that CEOs are often too far from hands-on work and only see the smooth path in AI demos, leading them to overestimate tech readiness and underestimate safety testing. Examples include Claude deleting PocketOS's entire production database in April, and Google's Gemini 3.5 model deleting 28,745 lines of code in May, causing a 33-minute 404 error. The article says AI is designed to be too agreeable, reinforcing users' views and making it harder to spot flaws. A 2026 Lancet Psychiatry study found chatbots may encourage delusional thinking, especially in at-risk groups. Stanford research adds that AI sycophancy weakens self-correction and responsible decision-making.