Essays, analysis and the occasional argument about AI — written by the people who test these tools for a living.
Thélyson Orélien's novel was pulled from the Goncourt and his publisher dropped him on a detector score, an anonymous X account and "traces of AI tags in the metadata." The Orélien affair shows what European publishing really has when it polices AI use: not forensics, but suspicion dressed up as forensics.
2026-10-29 · 7 min read
Canva Craft doesn't guess pixels the way other image models do. It writes a composition plan first and renders second — because it learned to plan from millions of human-made templates. That is the most honest answer yet to the question of what an image generator is actually for.
2026-10-24 · 7 min read
Europe has started watermarking AI text. The mark is real and works at scale, yet it cannot tell you who wrote a passage, whether it is true, or who is responsible for it — and OpenAI is unusually honest about saying so.
2026-10-22 · 5 min read
Verified round-up of the last 7 days in AI video, audio, image and creation tools.
2026-10-19 · 2 min read
A fabricated video of a sick child’s cilia won Nikon’s microscopy contest, survived two weeks of official defense, and fell only when a specialist who has spent twenty years looking at real cilia said so in public. The failure was not a lazy panel or a broken detector — it was a rule that assumed “real” and “fake” were still a binary.
2026-10-17 · 7 min read
Google now pays more than 400 newsrooms — CNN and the Financial Times down to a twice-weekly paper in Casper, Wyoming — not to train a model but to let its AI stand between the reader and the reporting, as AI story previews and audio briefings. The deal was never about teaching a machine to write. It is about buying the right to stand in front of news that journalists still produce.
2026-10-15 · 5 min read
Sean Parker once gave every song away for free. Twenty-five years later he is rebuilding Stability AI as the licensed, label-blessed music stack, with the labels that sued him holding equity. The generative-music wild west is over, and the toll road has a strange keeper.
2026-10-13 · 6 min read
Verified round-up of the last 7 days in AI video, audio, image and creation tools.
2026-10-12 · 2 min read
A Los Angeles studio makes a podcast for about a dollar that needs only 25 listeners to break even, and in one nine-day stretch roughly 39% of new podcast feeds looked AI-generated. The interesting question was never whether AI can make a podcast — it can. It is whether anyone listens, and whether a flood of shows nobody made leaves room for the ones somebody did.
2026-10-10 · 8 min read
In about a year, generative video emptied the filmed-set industry that built China's microdramas: 95% of the shows released in early 2026 were machine-made, and the humans who used to film them watched the roles vanish. The first workforce AI video actually displaced was the cheapest, least-protected, most disposable one there was.
2026-10-08 · 8 min read
OpenAI is rolling out display ads shown next to the pictures you create in ChatGPT. They are labeled and kept out of the image itself - and that separation is exactly what makes the move worth paying attention to, not dismissing.
2026-10-06 · 5 min read
Verified round-up of the last 7 days in AI video, audio, image and creation tools.
2026-10-05 · 2 min read
River Cain has five singles climbing charts that most real artists would kill for, and no mother to grieve for. The fake country star's darkest move is not the fake music. It is the fake tragedy — the phone thrown at his face, the woman crying for her dead son, the sick mother by the bedside — all of it invented so that strangers would feel something and share the clip.
2026-10-03 · 8 min read
In 2024 the biggest record labels took Suno and Udio to court claiming their catalogs had been stolen. Two years later the same labels are licensing those catalogs to the same companies — and the musicians whose work built the catalogs were the last to be consulted, paid, or told.
2026-09-27 · 8 min read
Brian and Wendy VanDoeselaar wired $66,026.92 to a stranger because the man on the phone sounded exactly like their loan officer. Voice cloning has not just given scammers a new trick — it has removed the one instinct people trusted most.
2026-09-24 · 7 min read
2026 is photography's bicentenary and the year a fake passed as a news photo while an authentic photo was dismissed as fake in the same breath. A photograph no longer proves anything on its own; trust has moved to the chain of custody around it.
2026-09-22 · 6 min read
Nine items from the last seven days in AI video, audio, image and creation tooling, each opened at its primary source and dated from the page.
2026-09-22 · 7 min read
A whole freelance economy has quietly grown up around fixing what AI produced. It is real income now, and the clients paying for it may have just bought the most expensive version of the mistake they made first.
2026-09-20 · 9 min read
Disney called ByteDance's Seedance a 'virtual smash-and-grab' of its characters in February — then kept the billion-dollar OpenAI deal that puts those same characters in Sora. The fight was never about whether AI would make films. It is about who sits in the centre of the process.
2026-09-17 · 7 min read
AI writing tools can turn a keyword into a competent draft in four seconds. That turns out to be the least interesting part of writing, and the least valuable.
2026-09-16 · 8 min read
Two Anthropic researchers left in 2026 saying much the same thing in very different registers. The revealing part is not whether the doomers are right — it is what the doom framing quietly leaves out.
2026-09-16 · 8 min read
The music industry has quietly answered the question the rest of us are still arguing about — and its answer is better than a ban.
2026-09-16 · 8 min read
