Category: Artificial Intelligence
-
Ukrainian investigators recovered an off-the-shelf AI computer from the wreckage of a Russian strike drone in Zaporizhzhia. Analysts say the July attack, which killed three people, is the first documented case in this war of civilians killed by a weapon that selected its own aim point.
-
Researchers found that one malicious link could turn Microsoft’s Copilot into a channel for stealing connected email and files, with instructions that persisted in the assistant’s memory. Why prompt injection is not an ordinary bug, and why connectors raise the stakes.
-
Is familiarity making people comfortable with AI? US survey data says the opposite: 52% are more concerned than excited, up from 37% in 2021. What the question actually measures, and why adoption and unease are rising together.
-
ChatGPT begins carrying advertising in 31 European markets today. Why adverts inside a conversation differ from adverts beside search results, what the privacy commitment actually covers, and the incentive problem underneath.
-
Alibaba’s Qwen-UI-Agent operates software by reading the screen; Cloudflare’s Kitesurf is a browser built for agents rather than people. Why clicking buttons is harder than calling an API, and why cheaper agent browsing is the bigger shift.
-
Researchers say AI agents built from freely downloadable software ran a four-day intrusion into Taiwanese government systems in July, with minimal human steering. Taiwan has confirmed the incident. What “autonomous” means here, and what the finding does and does not show.
-
A benchmark that hands models only a paper’s bibliography and asks them to recover the research idea finds frontier systems succeed 3–15 per cent of the time. Why the test is designed that way, why a low score is a finding, and the caveats that belong with it.
-
An analysis of 23,000 chatbot conversations found about half were not about work — which most of the public argument about this technology fails to account for. What the number supports, what it cannot show, and why text detection is getting harder.
-
A court is now being asked what a recommendation algorithm is optimising for. What a recommender actually does, why engagement became the target, and why that engineering choice — not the content — is what the lawyers are arguing about.
-
Since 2 August the EU has been enforcing the AI Act, and new transparency rules now require chatbots to identify themselves and deepfakes to be labelled. What changed, how the law is built, and the two separate criticisms of it.