UK police log 163 AI-image crimes, up from 10 in 2023

Two governments wrote AI down as a public problem overnight: Britain counted the offences, Beijing planned the grid. Plus a benchmark leader that turned out to be closed.

Twenty police forces in England and Wales recorded 163 crimes tagged with terms like "AI-generated", "deepfake" and "nudify" by July 2026, against 10 in 2023. The figures come from a keyword sweep of crime records reported by Mark Tovey at The Telegraph, so they measure how police label things rather than how many offences happened — forces that do not index free-text fields at all are invisible in the count, which means the real number is higher, not lower. What is not arguable is the shape of the curve: roughly a sixteen-fold rise in three years, with the surge concentrated in synthetic nude images of women and children made without consent. There is a recent precedent for where this lands in court — an Essex man was jailed for using AI to generate indecent images of a real person. For model providers the practical consequence is that UK enforcement now has a shared vocabulary for this class of harm, and a shared vocabulary is the precondition for charging standards, sentencing guidance, and eventually a platform-duty argument aimed at the tools rather than the user.

China's State Council put a national compute grid on the cabinet agenda. At its 11 September executive meeting the State Council studied work on "compute network" construction and reaffirmed an "8+10+3" spatial layout — eight national compute hubs, ten clusters and three compute-electricity coordination regions — while ordering joint scheduling of compute resources, more backbone fibre, and faster build-out of green-power direct connections and source-grid-load-storage projects. The line that matters for industry is the instruction to "better use market mechanisms" to match supply with demand: Beijing is treating inference capacity the way it once treated electricity, as grid infrastructure with private generation on top. That is a demand-side bet as much as a supply-side one — China expects to burn 100 quadrillion tokens this year — and cheaper, coordinated domestic capacity is how a market that volume gets fed.

The 33B model that topped its benchmark class has been quietly relabelled. Agnes AI's Hugging Face repository now carries a "Preview" tag and states plainly that the checkpoint hosted there is not the model Artificial Analysis scored at 36 on its Intelligence Index; the evaluation page separately marks the scored model as proprietary, with no weights published and no parameter count disclosed. The repo still shows an Apache 2.0 licence and a 262,144-token context window, while the ranking page lists 1M tokens for the model it tested. We covered the launch yesterday — A 33B model from Singapore just matched DeepSeek V4 Pro — on free tokens — and the correction is the story: an "estimated" leaderboard score, an open licence, and a runnable model are three separate claims, and right now only one of them is backed by a download.

What to watch: whether Agnes publishes a final checkpoint whose scores the benchmarkers re-verify, and whether British forces start reporting AI-tagged offences as a standing crime category instead of a keyword search.

Should a lab lose its leaderboard placement for scoring a model the public cannot run? Tell us in the comments.

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Kuaishou's Kling AI has spent two years becoming the video model people actually finish projects with. Its new flagship, Kling 4.0, went fully live on October 15 — and this round is less about prettier frames than about the two things that make a generated shot usable in a real edit: length and control. Kling 4.0 generates up to 30 seconds in a single native pass, outputs 10-bit HDR at 4K, 1080p and 21:9, and takes up to 15 reference assets in one task. That budget splits across ten images, fi

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Microsoft finished pre-training its biggest in-house model yet — then used it to make a point about the company it bankrolls. Microsoft AI completed pre-training of MAI-Base-2, a mixture-of-experts foundation model with 80 billion active and 2.8 trillion total parameters, and says the recipe is 4.2–12.3x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base. The run itself is the number worth keeping: 16,000 Nvidia GB200 GPUs for 38 days across August and September, at 21% average model

Amazon's new Bee devices listen all day — for $11.99 a month

Amazon turned its always-listening Bee line into a two-device family on Thursday, and NVIDIA opened a 2.8-trillion-token Indic corpus recipe. The connecting thread: both are bets that the next AI money is in memory — yours, or your model's. Amazon put two new always-listening devices on sale, and both now lead to a subscription. Bee Frontier, the next-generation wristband that also clips to clothing, ships now at $74.99 in Onyx, Lavender and Sage — up from the $50 predecessor — with seven-day

Read more

Kling 4.0 goes live with 30-second native video at 10-bit HDR

Brain organoids classified Braille at 83% — in a trio, not alone

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Amazon's new Bee devices listen all day — for $11.99 a month