Five things worth a second look this week.

01 · AI + PHOTOGRAPHY

The assault was real. The photo wasn't.

On 12 October, at a Trump rally in Youngstown, Ohio, a woman protesting in a shirt that read "Release the Epstein Files" was attacked. That part happened.

Then a picture of it started moving on Threads. It showed a crowd in red hats closing in around a figure in a black shirt. The post with it said a "brave woman" had been punched and dragged to the ground while police stood by.

The picture wasn't a photograph. Fact-checkers ran it through SynthID, the invisible watermark Google puts into images made with its AI tools, and it came back as AI-generated. The person who posted it then admitted making it. They called it a "symbolic graphic", meant to represent the event rather than show it.

So this isn't a fake event. It's a fake picture of a real one.

That's a different problem from the one we usually worry about. The fear with AI images has been that people will believe things that never happened. Here, something did happen. Someone decided the real record wasn't enough and made a cleaner, more shareable version.

The cost doesn't fall on the person who made it. It falls on the woman who was hurt. Once a fake is caught, it becomes part of the story. Anyone who wants to doubt the attack now has something to point to, and every real account of that afternoon has to compete with the debunk.

One more detail is worth noticing. The image was caught because the company whose model made it had watermarked it. Detection worked here because the maker labelled its own output. Not every tool does.

A drawing used to look like a drawing, and nobody took a cartoon for evidence. Now a "symbolic" picture can look exactly like a photograph. Both things are true in Youngstown: the woman was hurt, and the picture of it was made up. Which one people remember may depend on which one they saw first.

02 · TECH + CREATIVE SOFTWARE

Seven free copies of Adobe's apps. Built because of a cancellation fee.

Brandon Thomas, who runs ArtCraft, has released seven free, open-source apps meant to replace Adobe's. There are versions for Photoshop, Illustrator, Premiere Pro, Lightroom, After Effects, InDesign and Acrobat. He built them with Claude Opus 5.5, wrote them in Rust, and released them under the MIT and Apache licences.

The headlines were about AI rebuilding Creative Cloud. The apps themselves are early alpha builds. ArtCraft's own roadmap puts PhotoCraft at roughly 25 to 35% of what a professional needs for daily work.

The more interesting part is why he did it. He said he'd been caught by Adobe's cancellation fee one too many times.

Not "AI can replace Photoshop". "People are tired of renting the tools they work with."

03 · TECH + CREATOR ECONOMY

The US government now has a box for creators to tick.

On 15 October, the US Small Business Administration said creator-led businesses belong under NAICS 5121, the industry code for motion picture and video, when they apply for its programmes.

Creators could already get SBA support if they qualified. The problem was that nobody knew which box to tick. A podcaster, a streamer and a photographer running a YouTube channel didn't fit any existing category, and loan forms run on categories.

It isn't the end of the process. A related rule on size standards is still open for comments until 20 November, and nothing changes on tax. SBA Administrator Kelly Loeffler called it the first time the agency had recognised creators within its small-business framework.

It's a US change, but the point travels. Being a business partly means being legible to a bank's paperwork.

04 · AI + EDUCATION

AI gave teachers time back.

On 14 October, Gallup published a survey of 1,952 US school educators. 59% had used AI tools over the last school year. Of those, 64% said it gave them more time for the parts of teaching they value most. 29% saw no difference, and 7% said it cost them time.

The question wasn't how much more they produced. It was whether they got time back for the work they care about.

That's a better test of AI than most of the ones we use.

05 · SOCIAL + PLATFORMS

Substack is handing readers the dials.

On 14 October, Substack added content preferences to its feed. You can choose how much video, audio, longform writing and community posts you see, mute keywords and phrases, pick the topics you follow, and have AI-generated material labelled automatically.

Most feeds decide the mix for you and let you tweak it by scrolling past things. Here the reader sets it directly. Labelling AI content has become a setting you switch on, not a promise the platform makes on your behalf.

Whether people actually use the dials is the next thing to watch.