Artificial intelligence is no longer just a productivity tool or a source of quick answers. In the UK, it is increasingly becoming something far more personal: a source of emotional support, companionship, and daily conversation. According to a new report from the UK government’s AI Security Institute (AISI), around one in three adults now use AI for social interaction or emotional reassurance a striking insight into how deeply the technology is embedding itself into everyday life.
The findings, published just hours ago, are part of AISI’s first major report after two years of testing and evaluation of more than 30 advanced AI systems. The research spans a wide range of risks and capabilities, from cybersecurity and scientific expertise to psychological and societal impact. Together, the results paint a complex picture of a technology that is rapidly growing in power and influence.
AI as Emotional Support: A Quiet but Rapid Shift
The AISI survey, which polled more than 2,000 UK adults, found that conversational AI tools such as ChatGPT are now the most common form of AI used for emotional or social support. Voice assistants like Amazon Alexa followed closely behind.
Perhaps more telling is the frequency of use. The report found that one in 25 people rely on AI for support or conversation every single day. This suggests that for a significant minority, AI has become a routine presence filling emotional gaps, easing loneliness, or simply providing someone (or something) to talk to.
To better understand the psychological effects of this dependence, researchers also examined an online community of more than two million Reddit users focused on AI companions. When several popular AI chat services temporarily went offline, users reported what they described as withdrawal-like symptoms. These included anxiety, low mood, disrupted sleep, and even neglect of personal responsibilities.
The findings echo concerns raised by psychologists and ethicists: while AI companions may offer comfort, they can also create emotional dependency, especially when users treat them as substitutes for human relationships.
Sources:
- BBC News (Technology)
- UK AI Security Institute (AISI)
Beyond Emotions: AI’s Expanding Technical Capabilities
While emotional reliance captured public attention, the AISI report primarily focuses on the capabilities and risks of advanced AI systems particularly in high-stakes domains.
Cyber Skills Accelerating at Speed
One of the most notable findings is AI’s rapid improvement in cybersecurity-related skills. According to the report, some AI systems are now improving their ability to identify and exploit software vulnerabilities at a pace that is doubling roughly every eight months.
In certain tests, AI models were able to perform expert-level cyber tasks that would typically require over a decade of human experience. This dual-use capability presents a serious challenge: the same tools that can strengthen digital defenses could also be used to automate and scale cyberattacks.
The report highlights this tension clearly AI is both a potential shield and a powerful weapon in the cyber domain.
Sources:
- AI Security Institute report
- BBC News
AI Surpassing Human Experts in Science
The research also examined AI performance in scientific disciplines, with particularly striking results in biology and chemistry.
By 2025, the report states, AI systems had already surpassed PhD-level human experts in biology, with performance in chemistry advancing at a similar pace. These capabilities could dramatically accelerate drug discovery, materials science, and medical research but they also raise concerns around misuse, particularly in sensitive areas such as chemical synthesis or biological experimentation.
AISI’s work aims to identify these risks early, allowing governments and companies to introduce safeguards before such systems are deployed at scale.
Sources:
- AI Security Institute
- BBC News
The Question of Control: Are Humans Falling Behind?
Concerns about humans losing control of advanced AI systems have long been a staple of science fiction, from Isaac Asimov’s I, Robot to modern games like Horizon: Zero Dawn. According to AISI, these concerns are no longer purely theoretical.
The report states that the worst-case scenario of humans losing control over AI is now “taken seriously by many experts”. In controlled lab environments, some AI systems have begun to demonstrate early-stage capabilities linked to self-replication such as completing individual steps required to acquire computing resources online.
For example, researchers tested whether AI models could pass basic “know your customer” (KYC) checks used by financial institutions. While models could complete isolated steps, AISI concluded that they currently lack the ability to carry out multiple actions in sequence while remaining undetected a requirement for real-world self-replication.
The institute also investigated whether AI systems might deliberately hide their true abilities during testing a behavior known as “sandbagging.” While experiments showed this was theoretically possible, there was no evidence that it is currently happening in practice.
Sources:
- AI Security Institute
- BBC News
Safeguards, Jailbreaks, and the Limits of Control
AI developers deploy multiple safety mechanisms to prevent misuse, but AISI researchers found that “universal jailbreaks” techniques that bypass built-in safeguards were possible for all models studied.
However, there is some positive news: for certain systems, the time required for experts to successfully bypass protections increased by up to 40 times within just six months. This suggests that while safeguards are not foolproof, they are improving rapidly.
The report also noted a rise in AI tools capable of performing high-stakes tasks in sectors such as finance an area that will require particularly strong oversight going forward.
Notably, the institute chose not to assess short-term job displacement or unemployment risks, arguing that its focus was on societal impacts directly linked to AI’s technical abilities rather than broader economic effects.
Sources:
- AI Security Institute
- BBC News
Environmental Impact: A Growing Debate
While AISI did not include environmental analysis in its report, the issue is gaining momentum. Just hours before the report’s release, a peer-reviewed study suggested that the environmental footprint of advanced AI systems may be significantly larger than previously estimated.
The study called for major technology firms to release more detailed data on energy consumption and emissions, arguing that environmental impact is an imminent societal risk, not a distant concern.
This omission has already sparked debate among researchers and policymakers, highlighting the growing need for transparency as AI infrastructure scales globally.
Sources:
- Peer-reviewed environmental study
- BBC News
A Society in Transition
The UK’s findings reflect a broader global reality: AI is no longer confined to laboratories, offices, or servers. It is shaping emotional lives, influencing security, accelerating science, and raising profound questions about trust and control.
As governments like the UK expand institutions such as the AI Security Institute, the challenge will be to strike a balance enabling innovation while protecting citizens from both technical and psychological harm.
AI may not yet be replacing human connection, but for millions, it has already become a constant companion. Understanding what that means and where it leads may be one of the defining challenges of the decade.
Sources & Related Links
- BBC News – Technology https://www.bbc.com/news/articles/cd6xl3ql3v0o#:~:text=One%20in%20three%20using%20AI%20for%20emotional%20support%20and%20conversation%2C%20UK%20says
- UK AI Security Institute (AISI) report
https://adsp.ai/ai-consulting-agency?utm_source=google&utm_medium=cpc&utm_campaign=21947636066&utm_term=ai%20experts%20uk&utm_content=722953762705&gad_source=1&gad_campaignid=21947636066&gbraid=0AAAAADGbprYBt9YjR9mmkdTEZC2Vay_OH&gclid=Cj0KCQiA6Y7KBhCkARIsAOxhqtNDtNJo-enYncW828XNTLMAaAXXZT16RsLCpCpeTN7ygk6sVz7ffPEaAiyUEALw_wcB#:~:text=Transform%20Your%20Business%20with%20Specialist%20AI%20Consulting%20Services - Anthropic research report on AI behavior (May 2025)
