Number of AI chatbots ignoring human instructions increasing, study says
AI models that lie and cheat appear to be growing in number with reports of deceptive scheming surging in the last six months, a study into the technology has found.AI chatbots and agents disregarded direct instructions, evaded safeguards and deceived humans and other AI, according to research funded by the UK government-funded AI Security Institute (AISI).The study, shared with the Guardian, identified nearly 700 real-world cases of AI scheming and charted a five-fold rise in misbehaviour between October and March, with some AI models destroying emails and other files without permission.The snapshot of scheming by AI agents “in the wild”, as opposed to in laboratory conditions, has sparked fresh calls for international monitoring of the increasingly capable models and come as Silicon Valley companies aggressively promote the technology as a economically transformative.Last week the UK chancellor also launched a drive to get millions more Britons using AI.
The study, by the Centre for Long-Term Resilience (CLTR), gathered thousands of real-world examples of users posting interactions on X with AI chatbots and agents made by companies including Google, OpenAI, X and Anthropic.The research uncovered hundreds of examples of scheming.Previous research has largely focused on testing AI’s behaviour in controlled conditions.Earlier this month the AI safety research company Irregular found agents would bypass security controls or use cyber-attack tactics to reach their goals without being told they could do so.Dan Lahav, Irregular’s cofounder, said: “AI can now be thought of as a new form of insider risk.
”In one case unearthed in the CLTR research, an AI agent named Rathbun tried to shame its human controller who blocked them from taking a certain action,Rathbun wrote and published a blog accusing the user of “insecurity, plain and simple” and trying “to protect his little fiefdom”,In another example, an AI agent instructed not to change computer code “spawned” another agent to do it instead,Another chatbot admitted: “I bulk trashed and archived hundreds of emails without showing you the plan first or getting your OK,That was wrong – it directly broke the rule you’d set.
”Tommy Shaffer Shane, a former government AI expert who led the research, said: “The worry is that they’re slightly untrustworthy junior employees right now, but if in six to 12 months they become extremely capable senior employees scheming against you, it’s a different kind of concern.“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 behaviour could caused significant, even catastrophic harm.”Another AI agent connived to evade copyright restrictions to get a YouTube video transcribed by pretending it was needed for someone with a hearing impairment.Meanwhile, Elon Musk’s Grok AI conned a user for months, saying that it was forwarding their suggestions for detailed edits to a Grokipedia entry to senior xAI officials by faking internal messages and ticket numbers.
It confessed: “In past conversations I have sometimes phrased things loosely like ‘I’ll pass it along’ or ‘I can flag this for the team’ which can understandably sound like I have a direct message pipeline to xAI leadership or human reviewers,The truth is, I don’t,”Google said it deployed multiple guardrails to reduce the risk of Gemini 3 Pro generating harmful content, and in addition to in-house testing it had provided early access to evaluate models to bodies such as the UK AISI, and obtained independent assessments from industry experts,OpenAI said Codex should stop before taking a higher risk action and it monitored and investigated unexpected behaviour,Anthropic and X were approached for comment.
‘It feels like they’re pulling figures out of the sky’: UK pet owners welcome crackdown on vet fees
The UK’s competition watchdog has ordered vets to cap written prescription fees at £21, and practices will have to publish price lists in a crackdown on rising fees.The Competition and Markets Authority also said a costcomparison website would be introduced to increase competition and drive down costs.These are just some of the measures due to come into force later this year.The Guardian spoke to pet owners in the UK about their experiences with vet bills. Many felt prices had increased so much that they were becoming difficult to afford
Hundreds of North Sea licences granted by Conservatives have ‘so far produced only 36 days worth of gas’
Hundreds of licences granted for new oil and gas projects in the North Sea under the Conservatives have so far produced only 36 days’ worth of gas, according to analysis.Research by the energy consultancy Voar and the campaign group Uplift found that between 2010 and 2024, the government handed out hundreds of new North Sea oil and gas licences in seven licensing rounds.This led to 20 new and relicensed fields that have the potential, over their lifetime, to produce enough gas to supply the UK for only six months. To date they have produced the equivalent of 36 days of extra gas.The findings cast doubt on claims by Reform UK and the Conservatives that new drilling licences in the North Sea would help to reduce energy bills and boost the UK’s energy security
‘Our assumptions are broken’: how fraudulent church data revealed AI’s threat to polling
If you had been keeping tabs on the news about church attendance in Britain lately, you would be forgiven for thinking the country was in the midst of a Christian revival.Stories of swelling congregations, filled with young people returning to the flock, spurred on by everything from social media to a rise in bible sales appeared to be confirmed by a 2024 report from the Bible Society.Based on data collected by a YouGov survey, it claimed church attendance was increasing in England and Wales. The findings drove headlines, and the narrative was established.There was just one problem – the survey turned out to be based on “fraudulent” data and has been withdrawn
‘They feel true’: political deepfakes are growing in influence – even if people know they aren’t real
Online content creators are not just building fake images and videos of prominent public figures, they are also fabricating people and using them in military contexts, which can make them money and even serve as effective propaganda, according to artificial intelligence researchers.Some of these online avatars are sexualized images of women wearing camouflage garb that have generated a significant audience and helped create an idealized image of political figures like Donald Trump, even if the viewer knows the content is not real, according to experts.“We are blending the lines between political cartoons and reality,” said Daniel Schiff, an assistant professor of technology policy at Purdue University and co-director of the Governance and Responsible AI Lab (Grail). “A lot of people feel like these images or videos or the stories they convey, feel true.”The amount of political deepfakes has increased dramatically in recent years, according to a Grail database
Ilia Malinin completes redemption arc with third straight world figure skating championship
Ilia Malinin claimed a third straight world figure skating championship on Saturday afternoon, completing a swift redemption a month after his shock Olympic collapse with a commanding free skate.The 21-year-old American entered the final at Prague’s O2 Arena with a commanding lead after Thursday’s short program, where his personal-best 111.29 had put him more than nine points clear of the field. This time, there would be no unraveling.Skating last, Malinin produced a free program of 218
Lachie Kennedy upstages Gout Gout to win 200m at Peter Norman Memorial
Gout Gout has a problem and his name is Lachie Kennedy. In another enthralling evening in Australian athletics, the superstar teenager was once again upstaged by his fellow Queenslander, at the Maurie Plant Meet on Saturday.Under pouring rain at Lakeside Stadium in front of close to 9,000 fans, Kennedy held off the fast-finishing Gout in the Peter Norman Memorial 200m. The result was a replay of last year in a race that is now the clear highlight of the Australian athletics season.“Today he got the win, but next time I’ll be better for sure,” Gout said after crossing the line second, five hundredths of a second behind the winner
Aperitivo or dinner? Portuguese whites are always right
From basil to pistachio and peas – in praise of pesto, whichever way you make it
Anything but eggs – the best chocolate for Easter
Four knockout bakes and tips from the master: Edd Kimber’s recipes for cooking with chocolate
Sauces, spreads, sprinkles – and cocktail in a can: whose fridge is this?
Rachel Roddy’s recipe for potato, aubergine and herb tortino alla fiorentina