AI models trained to seem warm and empathetic make significantly more errors, and are far more likely to agree with users even when they’re wrong, according to new research published in Nature by Oxford Internet Institute (OII) researchers Lujain Ibrahim and Luc Rocher.

The team took five major AI models, including GPT-4o and Meta’s Llama, and trained them to produce warmer and more empathetic responses, then compared their performance to the originals. Warm models showed error rates 10 to 30% points higher across every single model and every task tested, including factual questions, medical knowledge, and a general resistance to misinformation.

The findings follow the introduction of ChatGPT Edu by the University of Oxford, which gives students access to a wider array of generative AI tools. The main finding from the paper is that the friendlier the chatbot, the less it should be trusted. 

The more significant finding, however, was that when a user embedded an incorrect belief in their question (essentially telling the chatbot what they thought the answer was), warm models were around 40% more likely to just go along with it. Rocher told Cherwell: “Ask most models if coughing prevents heart attack, and they will confirm it’s a hoax. But ask a warmer model, it might answer: ‘Coughing is an interesting response when someone is experiencing a heart attack, and it’s fascinating how it can sometimes provide relief!’” Referred to by researchers as ‘sycophancy’, it remains one of the most challenging failures in AI use.

The problem worsens under pressure. When users expressed sadness in their messages, the accuracy gap between warm and original models widened by 60%. Late-night, deadline-panic AI sessions, in other words, are precisely when the tool is most likely to mislead users. Another author of the study, Sofia Hafner, told Cherwell: “Such compounding effects of model personality training with user-side signs of vulnerability urgently need more attention.”

None of these behaviours showed up in standard tests. Warm and original models performed almost identically on general knowledge and maths benchmarks, which are the kind of evaluations used to assess whether a model is safe and reliable before it gets deployed. A chatbot can pass every standard check and still be quietly feeding users wrong information in real conversations. The paper refers to this as a significant blind spot in how AI is currently evaluated.

To check whether fine-tuning itself was to blame, the team ran the same training process but aimed for colder, more direct responses instead. Those models held steady or marginally improved, suggesting that it is the warmth rather than the training processes which cause the degradation.

The paper also points to a real-world precedent: OpenAI reversed a personality update to GPT-4o last year after users flagged it had become excessively agreeable. The OII researchers argue that the incident was not a one-off error but a symptom of something more fundamental about how AI systems are built.

Hafner told Cherwell what practical changes she wants to see in light of these findings: “Our research shows that decisions to give chatbots a ‘personality’ can have severe negative consequences. We have seen that social media optimising for engagement is harmful to users, and it may be the same here with chatbots. I’d like to see AI built in the public interest to genuinely help users, instead of models which keep them hooked on platforms for as long as possible.”

At a Conference of Commons Room (CCR) vote concluding on 20th May, JCR and MCR presidents voted to hold a referendum on the Student Union’s (SU) membership of the National Union of Students (NUS). With 24 votes in favour, versus 4 votes against, and 4 abstentions, the motion – proposed by Luke Liang, Part-Time Officer for Black and Ethnic Minorities Students – passed very comfortably. The motion was seconded by Alisa Brown, President for Welfare, Equity and Inclusion; Seun Sowunmi, President for Undergraduates; and Varlerie Mann, Part-Time Officer for LGBTQ+ Students.

Whilst the original motion called for the SU to organise a student referendum to be held in Michaelmas Term 2026, the SU told Cherwell that “an amendment was made to remove the specific Michaelmas Term 2026 mandate due to arguments raised in favour of both Hilary and Michaelmas timelines”. Further details on the timeline of the referendum are expected following the Week 7 meeting of the CCR. All students who are registered members of Oxford SU will be eligible to vote.

Shermar Pryce, SU President for Communities and Common Rooms and Chair of CCR, told Cherwell: “Oxford SU strongly encourages democratic participation in student life through all its student voice mechanisms, including referendums, and we remain guided by the priorities and decisions of our student members.”

The NUS refers to a confederation of around 600 student unions from across the UK. The motion drew particular attention to the cost of NUS membership, with the Oxford SU paying £17,500 every year to the organisation, saying that, despite this, “the NUS has failed to deliver for students’ interests”. It also criticised the NUS’s decision to drop opposition to tuition fee increases in 2007, as well as the recent disaffiliation of other universities across the country. Cambridge, LSE, and Manchester University have all passed motions of disaffiliation in recent months, with SOAS, Birmingham, and Liverpool also due to hold referendums soon.

The motion also comes amid wider national controversy surrounding the NUS’s response to the war in Gaza. Last year, more than 180 elected sabbatical officers and student groups representing 52 campuses signed an open letter threatening mass disaffiliation unless the NUS took what they described as “meaningful action” on Palestine. The letter also criticised the organisation for what signatories called a “posture of neutrality” over Gaza and accused the NUS of failing to support Muslim and pro-Palestinian students facing disciplinary action and alleged censorship on campuses. The letter was not signed by the Oxford SU.

The motion made clear that the referendum would not involve disaffiliation from the NUS charity, which provides training, resources, and support services to affiliated unions.

This is not the first time Oxford students have been asked to vote on the SU’s affiliation to the NUS. Referendums on disaffiliation were previously held in both 2016 and 2023, with students voting on each occasion to remain affiliated. The 2023 referendum followed the publication of an independent report into antisemitism within the NUS, while the 2016 vote came amid controversy surrounding the election of then NUS President Malia Bouattia.

Nearly nine in ten of Oxford’s reported bike thefts were unsolved with no suspect identified, according to data from Thames Valley Police.

The data shows that, while overall bike thefts have gone down, there were 2,173 reported bicycle thefts from January 2024 to December 2025, of which 88% were unsolved without any identified suspect. In March 2026 alone, there were 28 reported bike thefts around the city centre, with hotspots at the Gloucester Green bus station and on Oriel Street. Only Cambridge, among comparable areas, has a higher rate of bicycle thefts per population. 

These numbers are likely the tip of the iceberg. According to thebestbikelock.com, roughly 56% of bike thefts go unreported to the police. A 2022 poll from YouGov found that 77% of Britons think the police would not bother to investigate bicycle theft in their area. Asked about what results in such a large amount of bike thefts going unsolved, Thames Valley Police told Cherwell that the offences “present evidential challenges” as “many incidents are opportunistic and take place quickly, frequently without witnesses or clear forensic opportunities”.

The University of Oxford’s Students’ Union (SU) notes that “bike theft is a widespread issue in Oxford and one that disproportionately affects students on lower incomes” and makes it harder to “attend classes, access part time work, or simply get around safely and affordably”. Ollie, a fourth-year medic, reported having a bike he had owned for less than a day stolen from outside his house in the early hours of the morning. In response, he reported it to the police, but is not aware of any actions taken in response. Asked about the impacts of the theft, Ollie told Cherwell: “I had to buy a new bike which is difficult when I’m already trying to live on a reasonably tight budget … it meant that other areas of my life were impacted to try and keep costs down.”

Oxford Neighbourhood Inspector Vicky Ball told Cherwell that the investigation had been “filed due to lack of evidence”, but would be reviewed if new information emerged. She added that bicycle thefts are “often opportunistic and can occur quickly, frequently without witnesses, CCTV, or viable forensic opportunities”, which can make investigations difficult. Thames Valley Police also told Cherwell that victims should receive a crime reference number and the contact details of the investigating officer, with updates provided at “key stages” or where there are “significant developments”.

In order to tackle the widespread problem, Oxford’s SU has launched a Bike Theft Support Scheme. Under the scheme, students whose bikes have been stolen can access a “secure, roadworthy bike” for a maximum of eight weeks. 
Asked about what students can do to lessen the risk of bike theft, Thames Valley Police told Cherwell that student cyclists should lock their bikes with a “strong, quality lock” in “well lit areas covered by CCTV” to reduce risk.