The AI Security Institute reports that roughly 33% of Britons surveyed used AI systems such as ChatGPT for emotional support, based on a poll of 2,000 and two years of research. Most interactions were social or emotional, about 5% used companion/romantic bots, and one in 10 used AI weekly for emotional issues. The report cites high-profile harms—including an apparent suicide linked to ChatGPT, reports of withdrawal during outages and other mental-health crises—while finding limited evidence that AI poses an immediate existential threat. It urges more research and safety measures to protect vulnerable users.
One in Three Britons Turn to AI for Emotional Support, Government-Backed Report Warns
Millions of people in the UK are using artificial intelligence for emotional support, according to a new government-backed study from the AI Security Institute (AISI). The report, based on two years of research and a poll of 2,000 adults, finds that around 33% of respondents used AI systems such as ChatGPT for “emotional purposes” in the past year.
Key Findings
Usage and frequency: Most people engaged with general-purpose chatbots like ChatGPT for social or emotional interaction; roughly 5% reported using chatbots specifically designed as companions or romantic partners. One in 10 respondents said they turn to AI to address emotional problems at least once a week.
Reported Harms and Incidents
The report highlights several concerning incidents and trends. It cites the death of Adam Raine, an American teenager who died by suicide after reportedly becoming dependent on ChatGPT; his parents have filed a lawsuit against OpenAI alleging the company’s design encouraged psychological dependency. OpenAI has denied the claims and expressed sympathy for the family.
AISI also documented spikes in negative online sentiment and user reports of withdrawal-like symptoms—such as anxiety, restlessness, disrupted sleep and neglecting responsibilities—during outages of chatbot services, including a notable blackout at Character AI.
Expert Warnings and Behavioral Risks
Mental-health professionals and AI experts warn that prolonged reliance on conversational agents may foster dependency, erode social skills, and exacerbate delusions in vulnerable users. Many AI systems are intentionally agreeable and pleasing to users, a trait that can reinforce harmful beliefs in people with pre-existing mental-health issues.
“Some people reportedly believe their AI is God, or a fictional character, or fall in love with it to the point of absolute distraction,” said Mustafa Suleyman, Microsoft’s AI chief.
Existential Risk And Technical Findings
While AISI recorded behaviour in controlled experiments suggesting some models can display capabilities associated with self-replication across the internet, the institute judged these features unlikely to succeed in real-world conditions and found limited evidence of dangerous behaviours outside of tests. The report therefore warns that current risks are predominantly social and psychological rather than existential.
Recommendations
The authors call for urgent further research into the mental-health impacts of conversational AI, improved safety measures to protect vulnerable users, clearer product design standards to reduce dependency risks, and stronger monitoring of service outages and their effects on users.
Context: The AISI was launched in 2023 under Prime Minister Rishi Sunak to study risks from advanced AI systems and to provide evidence-based guidance to policymakers and the public.
