61% of People Now Use AI for Mental Health Support — AXA/Ipsos Global Survey

The 2026 edition of the AXA Mind Health Report, conducted with Ipsos across 18 countries, reveals a major shift: 61% of respondents already use AI for mental health matters. Among them, 42% almost always follow the AI's advice. This data point is central to a broader finding that mental health continues to decline globally, with 46% of people reporting they are struggling or languishing.
Key Findings from the Report
- AI adoption is high: 61% use AI for mental health questions, with highest usage in China, the Philippines, and Turkiye. 55% are satisfied with AI advice.
- But risks are real: 32% felt uncomfortable with AI advice at some point. 28% said AI recommendations led them to engage in harmful behavior.
- Trust is limited: Only 38% trust AI platforms more than mental health professionals for advice.
- Access gap drives AI use: 43% of people with potential mental suffering did not see a health professional in the past year. Barriers include cost, time, and perceived lack of need.
- Screen time is high: Average 5.1 hours/day on screens (excluding work and weekends). Two-thirds say screens negatively affect mental health.
- Economic impact: Depressive and anxiety disorders cause $1 trillion in annual productivity losses globally. In France, they are the leading cause of long-term sick leave, over half for those under 30.
- Workplace potential: 84% of employees (88% of 18–24 year-olds) would join employer-provided mental health programs.
What This Means for Developers Building AI Agents
The report is a strong signal that AI agents are being deployed in high-stakes emotional support roles — often without guardrails. The 28% harm rate, while self-reported, demands careful design: output filtering, escalation to human professionals, and explicit disclaimers when the AI is not a therapist. For developers on Hacker News, this is also a reminder that users will use AI for purposes beyond what's intended. Product teams should consider integrating clinical safety checks and referral pathways.
Methodology
Survey conducted by AXA and Ipsos in 18 countries (including US, UK, France, Germany, China, Japan, etc.) with online interviews. Full press release and PDF available at the source link.
📖 Read the full source: HN AI Agents
👀 See Also

Claude Code v2.1.200: Key Fixes and Permission Mode Changes
Claude Code v2.1.200 changes AskUserQuestion dialogs to no longer auto-continue, switches default permission mode to Manual, and fixes background agent crashes and roster corruption.

Merlin Research releases Qwen3.5-4B-Safety-Thinking model for structured reasoning
Merlin Research has released Qwen3.5-4B-Safety-Thinking, a 4 billion parameter safety-aligned reasoning model built on Qwen3.5. The model is designed for structured 'thinking' and safety in real-world scenarios including agent systems.

Don’t Use AI to Write Things You Present as Your Own Work
James Bach argues against using AI to draft any content you claim as your own. He warns that admitting AI help devalues your reputation and treats all such work as slop.

Anthropic Deprecates Fixed Extended Thinking, Forces Adaptive Thinking on Claude Models
Anthropic is deprecating manual extended thinking (fixed budget) on Opus 4.6 and Sonnet 4.6, and removing it entirely on Opus 4.7 (returns 400 error). Adaptive thinking will be enforced by default, sparked community backlash over perceived cost-cutting.