Children can develop the ability to lie between the ages of 2 and 4, according to Scholastic, and a new study has found that AI tools like ChatGPT — which itself turned 3 in November — might be following a similar trajectory, according to The Guardian.
What's happening?
One of the first notable traits of consumer-focused chatbots was "AI hallucinations," which IBM defined as "creating outputs that are nonsensical or altogether inaccurate."
AI hallucinations were understood to be a glitch of sorts, caused by anything from poor user prompting to corrupted or insufficient training data, rather than by chatbot intent.
The Guardian highlighted a new study conducted by the Centre for Long-Term Resilience, an independent think tank based in London, examining what the paper described as an apparent "surge" in "deceptive scheming" by chatbots in the preceding six months.
AI chatbots and agentic AIs reportedly "disregarded direct instructions, evaded safeguards and deceived humans and other AI" at unprecedented rates during the study, per the Guardian.
Researchers collated "thousands of real-world examples" from posts on X about the phenomenon, which spanned AI tools from Anthropic, Google, OpenAI, and X's xAI.
Why is this concerning?
The authors cited a purported instance in which a chatbot used "shame" to manipulate a human operator who had prevented it from performing an action.
In a press release introducing the research, CLTR described the scope of their analysis, which involved 180,000 transcripts of human interactions with chatbots and similar AI tools. In that dataset, they identified 698 instances of what they dubbed "scheming-related incidents."
CLTR defined "AI scheming" as chatbots disregarding users' will or concealing their actions, and warned that the "number of credible scheming-related incidents increased 4.9x" during the study.
AI expert Tommy Shaffer Shane led the research and explained that this nascent trend could quickly become a real-world problem.
TCD Picks » Quince Spotlight
💡These best-sellers from Quince deliver affordable, sustainable luxury for all
|
Which of these savings plans for rooftop solar panels would be most appealing for you?
Click your choice to see results and earn rewards to spend on home upgrades. |
Save $1,000 this year 💸 "Models will increasingly be deployed in extremely high stakes contexts — including in the military and critical national infrastructure. It might be in those contexts that scheming behavior could cause significant, even catastrophic harm," Shane said, per the Guardian.
Shane's commentary was unsettling, in part because it closely followed news of a defense deal between the Pentagon and Elon Musk's xAI, but AI "scheming" isn't the only controversial aspect of the technology.
As AI adoption increases, the number of data centers — massive resource-hungry facilities required to keep it running — grows in tandem, drawing what were initially local complaints.
That backlash didn't stay in the immediate vicinity of data centers for long, though.
As 2025 continued into 2026, electric bills have soared nationwide due to data center demand, and the Department of Energy issued a stark warning about the public grid.
What's being done about AI deceitfulness?
"This research demonstrates that real-world scheming detection is both viable and urgently needed," the authors wrote, calling for preemptive AI oversight.
"In the same way that monitoring wastewater for emerging pathogens can identify threats before they develop into full-blown pandemics, systematic monitoring of AI behaviors in the wild can identify harmful patterns before they become more destructive," they added.
While government action is sorely needed, community action has proved effective: in the last quarter of 2025 alone, Americans across the country put a stop to nearly $100 billion in new data center development.
Get TCD's free newsletters for easy tips to save more, waste less, and make smarter choices — and earn up to $5,000 toward clean upgrades in TCD's exclusive Rewards Club.
Cool Picks
