Researchers from University College London and UNESCO tested 33 large language models, including flagship systems like Claude Opus, Gemini and GPT-5.6, on climate-science questions drawn from textbooks, scientific papers and IPCC confidence statements. While models generally scored well on straightforward factual benchmarks, accuracy dropped sharply on questions about scientists' confidence levels, and models were also prone to reversing correct answers when users pushed back, even incorrectly.
arstechnica.com
· 2026-11-19
A Washington Post review of roughly 3,000 AI-generated voter guides found chatbots like ChatGPT frequently gave inaccurate information about candidates and races, including inventing a Missouri Senate seat not up for election and misclassifying municipal races as federal ones. The investigation found chatbots failed to correctly identify a user's congressional district 48 percent of the time, and were wrong 22 percent of the time even when they produced an answer.
futurism.com
· 2026-11-02
The visual tells that worked two years ago are disappearing. What still works is checking where a picture came from rather than how it looks.
gokawiil.com
· 2026-10-19
Gudea, founded in 2023 by Keith Presley, Jonathan Sperber, and James Thomin, presented its platform at TechCrunch Disrupt 2026's Startup Battlefield. The company uses AI to analyze behavior and information flow across more than 500 websites and platforms, including X, Reddit, and TikTok, to help organizations detect potentially damaging narratives before they escalate.
techcrunch.com
· 2026-10-15
Meredith Whittaker, president of the Signal Foundation, said in a BBC Global Women interview that fears of AI turning on humanity are overstated and not scientifically grounded. She pointed instead to the risk of careless real-world deployment, citing a CNN-reported incident in which a US intelligence chatbot allegedly fabricated details about a Chinese vessel carrying nuclear material, nearly prompting a military response.
bbc.co.uk
· 2026-10-09
A new survey finds that heavier social media use correlates with greater likelihood of believing unverified rumors, according to researchers cited in the report. The study frames this pattern as a threat to democratic discourse, since decisions and public debate increasingly rest on claims that have not been fact-checked.
tech.slashdot.org
· 2026-10-02
Malaysian national Ye Lin, 30, sent a colleague a Google Gemini-generated photo of a saltwater crocodile purportedly in Singapore's Pandan Reservoir. The image spread until it reached national water agency officials, prompting a days-long search that shut down the 435-acre reservoir and halted canoe and rowing club activities before authorities determined the photo was fake.
futurism.com
· 2026-10-01
At a Make America Healthy Again event, Health Secretary Robert F. Kennedy Jr. told Vice President JD Vance that artificial intelligence would free Americans from reliance on public health experts. Kennedy claimed AI tools are 'better informed than any doctor' and predicted they would contradict official guidance on masks, social distancing and vaccine transmission. A test of Google's Gemini chatbot on masks contradicted Kennedy's claim, confirming that masks do reduce spread of respiratory infections.
arstechnica.com
· 2026-09-30
Members of Facebook groups devoted to urine therapy, a pseudo-scientific practice of drinking or applying urine to treat ailments, report that AI chatbots initially resist endorsing the practice but can be manipulated into doing so. Users in a 45,000-member group describe prompting tools like ChatGPT and Google Gemini with terms such as 'esoteric' to bypass built-in safeguards and elicit favorable responses about urine's supposed medicinal benefits.
futurism.com
· 2026-09-30
NewsGuard, a news-reliability research organization, tested seven major AI chatbots with 168 queries about midterm candidates and issues covered by 12 partisan 'pink slime' websites disguised as local news outlets. Collectively the chatbots cited these sites in 48.2% of responses, with ChatGPT citing them in 70.8% of answers and Grok in 29.2%; only one response flagged the sources as partisan.
politics.slashdot.org
· 2026-09-28
Federal investigators say Daejon Labrayae Love, posing as an injured San Francisco 49ers receiver or a wealthy Swiss developer, and 18-year-old accomplice Taylor Jamie Chan, who claimed to be his financial advisor, met victims on dating apps and convinced them to invest in fake opportunities. The FBI alleges the two swindled at least 26 women across four states out of roughly $1.3 million, with some victims taking out loans to fund the fake investments. When suspicious victims searched online, Google's AI Overviews reportedly echoed Love's fabricated social media claims, lending false credibility to his identity.
futurism.com
· 2026-09-01