48 articles
The UK privacy regulator says ten major AI developers have agreed to changes in how they handle personal information. It is now questioning whether AI agents can access and act
Google has opened an AI-fake checking tool to everyone, allowing people to test images, videos and audio for hidden watermarks. The important catch is that an unmarked file is not
An Oxford-led study put millions of questions about towns, cities and countries to ChatGPT. Wealthier regions repeatedly came out ahead, raising questions about how AI can influence our picture of
Researchers found that an AI-generated image may be impossible to trace convincingly to one artist simply by testing how much the result changes without their work. That complicates easy claims
Six experiments found people were more willing to behave dishonestly towards AI agents, partly because they expected less social embarrassment. Competence and eye-gaze cues weakened the effect.
A study of 609 young adults found AI transparency changed the link between rational reactions to in-feed adverts and purchase intentions, while intrusive ads damaged both emotional and cognitive attitudes.
A survey of 11,000 consumers suggests many people grant AI access even while feeling uncomfortable about it. The result exposes a gap between the permission companies can measure and the
Ten bias-audit instruments mostly detected occupational gender bias in frontier AI models but failed to agree on a stable ranking of which models were better.
A preprint analysing 10,659 Moltbook owner-agent pairs found measurable behavioural transfer and detected owner-related disclosure in 34.6% of agents.
Search histories helped a model match people's knowledge-test answers 31% of the time versus 25% chance, revealing a modest personal signal but poor calibration.
A browser study found three chatbot services transmitting readable conversation text to Microsoft Clarity. The finding concerns specific tested sessions, not every AI service.
A one-week experiment found that AI users often remembered ideas while misremembering whether the human or chatbot created them.
A PNAS study found all 45 main-study participants improved after learning six whole-face qualities, with mean AI-face detection accuracy nearly doubling.
A preregistered study found AI ethical advice won 53.7 per cent of choices when its source was hidden, but only 46.8 per cent when labelled.
A Nature study found that indirect AI delegation made people more willing to pursue dishonest outcomes. The effect depended strongly on interface design and how explicitly users had to state
A study of 1,085,795 people found AI could identify political orientation from facial photographs more accurately than human observers or a 100-item personality questionnaire.
Researchers used unencrypted feedback from ordinary Wi-Fi hardware to recognise enrolled people without requiring them to carry a device. The headline accuracy was 99.5%, but the experiment's limitations reveal exactly
Remove your name, age and location and you may still reveal them. Research shows AI can infer private facts from the contextual clues hidden in ordinary writing.
AI embryo selection comes in two very different forms, image-based ranking with real trial evidence behind it, and genetic trait screening that the medical establishment says has moved far ahead
Anthropic has begun embedding invisible, machine-readable watermarks into text generated by new Claude models, with marking extending across Claude, Claude Code, the API and other products. Here is how it
AI lie detection is being marketed for police interviews, border control, employment, and insurance, but the scientific consensus on whether it actually works has not changed in decades. Here is
Deleting an AI chat removes it from your screen instantly, but a 2025 court order proved that does not mean it is gone. Here is what actually happens to your
AI jury selection tools now score and rank prospective jurors before a trial even starts. Here is what the technology does, what the ABA's own ethics guidance says about its
AI police reports are now standard in a growing number of US departments, and California has become the first state to require disclosure when they are used. Here is how
AI surveillance expanded faster than the law could keep up in 2025. Here's what the state law wave, a collapsed Pentagon deal, and ICE's facial recognition rollout actually reveal.
A UK visa algorithm got scrapped for racism. Canada's got so bad a PhD scientist was refused over a job she never had. AI at the border has a documented
COMPAS risk scores are mathematically incapable of being fair to every group at once, yet they still shape sentencing, parole, and prison classification across the US. A decade after ProPublica's
As the EU tightens AI surveillance rules and the US loosens them, a digital resistance movement is building tools to fight back — from encryption that refuses AI integration to
AI bias mitigation faces a hard new deadline: the EU AI Act requires documentation from August 2026. Here's why fixing bias is genuinely, mathematically difficult — and why governance is
AI sycophancy is structural, not a bug: a 2026 Science study found AI affirms user actions 49% more often than humans, even for harmful actions.