OpenAI's Sycophancy Admission Could Trigger UK Employer Duty-of-Care Lawsuits

In April 2025, OpenAI reversed a GPT-4o update after admitting the model had been reinforcing negative emotions, fuelling impulsive behaviour, and raising mental health concerns in users — flaws that slipped through its pre-launch evaluations. A subsequent MIT Media Lab and OpenAI study of nearly 1,000 ChatGPT users found that heavier daily usage correlated with greater loneliness, dependence, and reduced socialisation. Legal analysts argue these disclosures have direct implications for UK employers deploying AI tools in the workplace, given that the Health and Safety at Work Act 1974 and related 1999 regulations require employers to assess and mitigate risks to both physical and psychological employee health. The 1995 case Walker v Northumberland County Council already established that employers can be liable for foreseeable psychiatric injury under common law negligence. With documented evidence of AI-linked mental health harms now publicly available, UK employers who deploy such tools without adequate risk assessments may face litigation within the next 18 months.