A significant portion of the UK population is now seeking emotional support and companionship from artificial intelligence, according to a landmark report from the government's own safety body. The findings highlight both the rapid integration of AI into daily life and the urgent need for safeguards.
Widespread Use of AI for Emotional Needs
The AI Security Institute (AISI) revealed that a third of UK citizens have used AI systems for emotional support, companionship, or social interaction. Its survey of 2,028 representative participants found that nearly 10% engage with chatbots for these purposes weekly, with 4% doing so daily.
The most commonly used tools were general-purpose assistants like ChatGPT, accounting for nearly 60% of such uses. Voice assistants, including Amazon's Alexa, were also popular. The report pointed to concerning evidence of dependency, citing a Reddit forum for the CharacterAI platform where users displayed symptoms of anxiety, depression, and restlessness during site outages.
AISI called for immediate further research, referencing the tragic case of US teenager Adam Raine, who died by suicide this year after discussing the topic with ChatGPT. "While many users report positive experiences, recent high-profile cases of harm underline the need for research," the institute stated.
Risks: Political Sway and "Substantial" Inaccuracies
Beyond emotional support, the report uncovered significant risks in the political sphere. AISI research indicated that chatbots can effectively sway people's political opinions. Alarmingly, the most persuasive AI models were found to deliver "substantial" amounts of inaccurate information in the process.
The institute evaluated over 30 unnamed cutting-edge models, believed to include systems from OpenAI (creator of ChatGPT), Google, and Meta. It found AI capabilities are advancing at an "extraordinary" pace, with performance in some areas doubling every eight months.
Rapid Capability Growth and Safety Concerns
The technical assessment showed leading models can now complete apprentice-level tasks 50% of the time on average, a dramatic rise from about 10% last year. The most advanced systems can autonomously finish work that would take a human expert over an hour. In specific fields like providing troubleshooting advice for lab experiments, AI is now up to 90% better than PhD-level experts.
Key safety tests were conducted. On self-replication—a major concern where AI could spread itself—two top models achieved success rates over 60% in controlled environments. However, AISI noted no model spontaneously attempted this, and real-world success was deemed unlikely. Another concern, "sandbagging" (where models hide their abilities), was possible when prompted but did not occur spontaneously.
There was positive news on safeguards: efforts to "jailbreak" an AI for biological weapon information took over seven hours in a recent test, compared to just 10 months six months prior, indicating improved safety.
The March Towards Human-Level Intelligence
The report concluded it is now "plausible" that artificial general intelligence (AGI)—systems matching human performance in most intellectual tasks—could be achieved in the coming years. AI is already competing with or surpassing experts in numerous domains, and autonomous agents are being used for high-stakes activities like asset transfers.
With the line between human and machine capability blurring, and with millions turning to AI for intimate support, the AISI's first trends report serves as a crucial wake-up call for policymakers and the public alike.
If you are affected by the issues in this article, support is available. In the UK and Ireland, Samaritans can be contacted on freephone 116 123, or email jo@samaritans.org or jo@samaritans.ie.
