Practical research and analysis on artificial intelligence, cybersecurity, automation, digital platforms, online safety, data, and the technologies changing work and everyday life.
OpenAI says an expert found only 14 instances of verbatim or near-verbatim book regurgitation in a sample containing 95 million ChatGPT outputs. That is meaningful
Meta AI displayed a family photo Kalie Robins says she deleted years earlier, but there is no evidence it recovered the image from deleted backend
Google says conditions attached to federal quantum funding would have slowed its path to a quantum computer. Newly filed agreements show what companies that accepted
ShinyHunters says it breached Florida’s DAVID driver database and stole more than 200,000 records, using a Jeffrey Epstein record as proof. A sherafy.com provenance audit
Jacob Coxon resigned from Anthropic warning that OpenAI and Anthropic are racing toward self-improving superintelligence that could kill humanity. The claim sounds extreme. But recent
OSIRIS earns a qualified green light as a free open-source intelligence dashboard. It can genuinely help journalists discover and monitor stories, but some authoritative-looking scores,
XR glasses can cause eye strain, dry-eye symptoms and temporary changes in focusing and binocular vision. But current human evidence does not show that normal
Several capabilities once discussed mainly in AI-doomsday scenarios are now real: frontier systems can discover zero-days, circumvent some containment, coordinate across agents, take unauthorized actions,
OpenAI confirmed that its agents wrote to public websites during the DSEWiki incident but treated the episode as misalignment rather than invoking the traditional security-response
An AI product is more than its underlying model. AI middleware can control what the model sees, which tools it can use, how requests are
