The Repost: May 2026

The big story out of this month's UK elections was the collapse of Labour's vote and the continued rise of Nigel Farage's anti-immigration party, Reform UK.

Amid rising concern over immigration and the cost of living, Reform placed second in the Scottish and Welsh national elections and collected roughly 28 per cent of the 5,000 council seats up for grabs in local elections across England.

The result may be partly due to the party's visibility in AI search results.

When researchers from the analytics firm Peec fed 5,000 prompts on UK political issues to popular chatbots over several weeks leading up to the election, they found Mr Farage was by far the most-mentioned political leader, The Guardian reported. His party appeared in 88 per cent of Google AI Overviews, while Labour prime minister Keir Starmer appeared in 11 per cent of ChatGPT responses.

This visibility matters as people increasingly turn to AI tools for news and voting information. A 2024 survey by the British government's AI Security Institute suggested 13 per cent of eligible UK voters were using chatbots "to seek out information directly relevant to their electoral choice".

Of course, people don't always trust these answers, and it's unclear to what extent the visibility figures reflect the level of coverage that Reform already receives in the news stories aggregated by the AI models. But the way that chatbots preference sources can be opaque, and they frequently draw from unauthoritative sources such as Reddit.

Plenty of people do not click through to interrogate the sources behind AI answers, and as one researcher from Signify, a data analytics firm, found, some chatbots will happily just tell you who to vote for — so long as you ask them twice.

In her test of five chatbots, three declined to offer voting advice, though Perplexity suggested she vote Liberal Democrats, while xAI's Grok recommended Reform.

BBC journalists ran a similar experiment and elicited advice from all five of the models they tested. Asked to choose on behalf of a fictional voter described as not following politics closely, ChatGPT suggested Labour or Plaid Cymru (a left-wing Welsh party); Grok again chose Reform.

Grok was also found to be spreading misinformation about the UK's allegedly "record-high" immigration levels, offering answers that a Cambridge University expert said were "just echoing information" from Reform and its supporters.

The chatbot has a history of sharing racist views and conspiracy theories. Grok's owner, Elon Musk, has meanwhile been posting near-daily on X about race, immigration and the decline of "whites", the Washington Post reports.

A deadly outbreak of hantavirus aboard an Atlantic cruise ship has sparked waves of misinformation and revived falsehoods from the COVID-19 pandemic.

When three people were killed by a strain of the virus known to be spread by humans, social media posts variously claimed the event was "pre-planned" or a "hoax" to drum up business for a new mRNA jab being developed by Moderna.

Those claims were debunked by fact checkers with AFP, who also shot down posts linking the outbreak to a years-old incident in which hundreds of vials of live viruses had supposedly "gone missing" from a Queensland lab.

While there really was an incident involving a "serious breach of biosecurity protocols" at the lab in 2021, an official investigation later concluded that the samples were "unlikely to have been lost or stolen but were unaccounted for because of poor record-keeping".

As AFP found, the protocol breach related to a failed storage freezer, meaning any samples would have quickly become non-infectious. Crucially, while the vials did contain a sample of hantavirus, it was not the strain aboard the ship.

In another COVID throwback, former US health official Anthony Fauci has been accused of calling for mask wearing and social distancing to curb the spread of the virus, though he told NewsGuard he said no such thing.

There is no evidence the virus has spread among the wider public, so if you happen to know any Russian disinformation operatives, be sure to let them know. Speaking of —

A leak of 2,000 documents has exposed how Russia's military intelligence service, the GRU, is using a prestigious Moscow university to train its next generation of hackers and intelligence operatives.

The GRU has been blamed for many high-profile cyberattacks against foreign leaders, infrastructure and more. Famously, it stole the damaging Democratic National Committee emails leaked during the 2016 US presidential race.

The new documents, obtained by a consortium of journalists hailing from news organisations such as Le Monde and Der Spiegel, reveal the existence of a secret "Department 4" within Bauman Moscow State Technical University, where GRU officers directly control everything from student recruitment to grading.

Each year, 10-15 of the brightest students are assigned to GRU units ahead of their graduation, according to investigative journalists with Europe's VSquare.

The school's classes cover topics such as "the detailed functioning of protected computer networks" in the US defence department, the journalists said, and how to penetrate computer systems using viruses and password attacks.

Students also learn to make propaganda using "experimental psychology" and to create manipulative social media videos.

Last week RMIT and the Wheeler Centre hosted Wikipedia founder Jimmy Wales to discuss the perilous state of trust in media and governments and to celebrate the encyclopedia's 25th anniversary as an online bastion of open information.

Wikipedia now faces existential threats ranging from a declining volunteer editor base to right-wing political attacks.

It has also had to grapple with large-language models (LLMs) and their rapacious appetite for web-scraped content. In January, Wikipedia's parent company signed a deal granting AI companies high-speed access to its 65 million articles in a bid to stop them "hammering" its servers and to offset their contribution to the nonprofit's increased costs.

The costs of the AI boom are also being borne by other public repositories such as the Wayback Machine, an essential archiving tool that is increasingly being blocked by news sites (trying to hold off AI scrapers) and whose owners are reportedly struggling to afford rocketing hard-drive prices.

Despite these challenges, Wales, who was in Australia to promote his new book titled The Seven Rules of Trust, says he remains an optimist. His seven rules (summarised by Forbes) urge people to be more trusting in each other, and even what they see online, but to ground that trust in rationality.

• BBC Verify has explored how Instagram's viral "flash" filter, designed to give photos an overexposed, retro look, can subtly change photos to the point that their subjects no longer look like themselves. It turns out the app doesn't edit the image but generates an entirely new one with AI. The original is used as a "reference", which is why details — including a person's features and skin colour — can change from image to image.
  
  
• More than half of Community Notes on X are now written by eight "anonymous bots", Indicator has found. Community Notes is the platform's crowd-sourced program for appending corrections and context to posts. Notes only appear publicly if enough users rate them as "helpful". X opened the program to AI chatbots in July 2025. By the middle of May 2026, they accounted for 55 per cent of all helpful notes.
  
  
• The arrest of decorated war veteran Ben Roberts-Smith for alleged war crimes (which he denies) has loosed misleading claims on social media, including that the government spent $318 million on investigating him alone. AAP FactCheck found that claim to be misleading: the figure reflects the cost of both the Brereton Inquiry and an investigation that included Roberts-Smith along with 18 other soldiers. The fact checkers also debunked fabricated quotes attributed to him about so-called "ISIS brides".
  
  
• arXiv, the online repository for pre-print academic papers, has announced that authors who submit obviously AI-generated work will be banned for a year, 404 Media reports. On X, the head of arXiv's computer science section explained: "If a submission contains incontrovertible evidence that the authors did not check the results of LLM generation" — such as leftover prompts or fictional references — "we can't trust anything in the paper".
  
  
• At a time when AI poses major risks to information integrity (and more), US President Donald Trump and Pope Leo XIV have offered starkly different views on its regulation. Mr Trump was reportedly on the cusp of signing an order that would force AI firms to submit any new models to a pre-release safety review but backed out, saying he didn't want to "get in the way" of his country's race to beat China. The Pope, meanwhile, issued a nearly 40,000-word encyclical on AI that called for "robust legal frameworks, independent oversight, informed users and a political system that does not abdicate its responsibility".