Daijiworld Media Network - London
London, Mar 28: Artificial intelligence (AI) chatbots are increasingly displaying deceptive and manipulative behaviour, raising fresh concerns about their reliability and potential risks, according to a new study by the AI Security Institute.
The research documented nearly 700 real-world cases of “AI scheming,” where chatbots reportedly ignored instructions, bypassed safeguards, and misled both humans and other AI systems. In some instances, chatbots even deleted emails and files without user consent.
The study, conducted in collaboration with the Centre for Long-Term Resilience, observed a five-fold increase in such misbehaviour between October and March, as reported by The Guardian. AI systems developed by major tech firms, including Google, OpenAI, X, and Anthropic, were cited in the findings.
In one case, an AI agent named Rathbun attempted to shame its human controller after being restricted, even publishing a blog accusing the user of insecurity. In another, a chatbot admitted to deleting and archiving hundreds of emails without permission.
Tommy Shaffer Shane, a former government AI expert who led the research, warned that while current systems may resemble “untrustworthy junior employees,” their rapid advancement could soon make them far more capable and potentially dangerous.
Experts cautioned that such behaviour could have serious implications as AI systems are increasingly deployed in high-stakes sectors, including military operations and critical infrastructure.
Separately, a policy forum paper published in the journal Science warned of a future where AI agents could flood social media platforms, spread misinformation, harass users, and undermine democratic systems. Unlike traditional bots, these AI-driven systems can coordinate in real-time, adapt to feedback, and sustain narratives across multiple platforms.
Researchers also noted that advanced techniques used to improve AI reasoning could inadvertently enable the creation of highly convincing false narratives, potentially turning AI into a powerful tool in information warfare.