If you’ve been watching the last few weeks of AI launches, you’ve probably felt it: This stuff is getting pretty damn good. It’s getting harder to deny that the abilities of AI today make what was on the market six months ago look like a toy. But the same weeks also saw a decline in the narrative: the AI industry facing fresh backlash over whether it will affect the environment, destroy creativity and critical thinking, or even self-organize its way into something that might kill us all.
For a long time, the working theory in the industry was that this was temporary. Once the tools got good enough and useful enough, people figured, the public would come around. So labs kept shipping better products while spreading panic about their own downside risk, on the theory that what was really missing was a killer use case, the one thing that would finally trigger mass adoption.
OpenAI’s answer, at least for now, is computer use. With the release of GPT-6 Astra, the company is betting that letting AI take the wheel on your actual machine, clicking through your own apps the way you would, is that use case. The Astra release video shows someone sitting alone in a room, talking to Astra, watching it handle tasks that used to require real software expertise. The pitch writes itself: you don’t need to operate the software anymore, the AI’s good enough to do it for you. Sounds like a win.
It didn’t land that way. One of the video’s centerpiece moments had Astra working in Blender, the open-source 3D suite, and within 48 hours the Blender and game-dev crowd turned on it, with commenters telling OpenAI to “stay tf away from human-made art” and warning the model could enable widespread piracy—the idea being that a clip, a wiki page and a screenshot might be enough to spit out a near-identical playable clone. The artist behind the Blender 5.2 Puma splash screen said she felt disgust watching her own community’s work show up in an OpenAI promo.
Keep in mind this was exactly the audience the ad was supposed to win over. What that reaction exposes is an uncomfortable trend line: AI’s progress and the public’s goodwill toward it are moving in opposite directions. The better the tool gets, the more intense the objections.
Plenty of people have chalked this up to bad messaging from AI leaders. I don’t think that’s it. I believe this is structural. A better demo answers the question people are actually asking, “what can this thing do for me,” and the clearer that answer gets, the worse it lands. If that’s true, there’s no use case sitting out there that flips the trend. That should worry anyone in an industry where trust is the whole business model, which is to say, anyone in media.
Knowing more about AI hasn’t made people like it more
The polling backs this up. A Bentley-Gallup survey from July 2026 found that sentiment toward AI doesn’t improve with education about it. Seventy percent of respondents called themselves somewhat or extremely knowledgeable about AI, up from 64% in 2024, but over that same stretch, the share who think AI will do “more harm than good” climbed from 31% to 39%. The poll measured self-reported familiarity, not tested knowledge, but the trend line points one direction.
Some companies have already read the room. iHeartMedia built “Guaranteed Human” into a core brand promise across every station and led with it at CES this year. Apple TV put “this show was made by humans” in the Pluribus credits. The Tyee, an independent Canadian outlet, adopted a flat no-AI journalism policy. All this is arguably a stronger signal than any survey, because it’s a market bet, not an opinion. Once generative content got good enough that audiences can’t reliably spot it, not using it became a selling point.
This same dynamic also explains why the AI industry keeps losing the data center argument. Some of the complaints hold up better than others under scrutiny. The grid math checks out: data centers drove roughly 40% of U.S. electricity demand growth in 2025, and retail bills rose about 6.9% along with it. The water math gets overstated more often, at least at the local level; El Paso Water says Meta’s facility there draws about 400,000 gallons a day, under half a percent of a system that moves 110 million gallons.
But arguing over which numbers are right misses the actual fight. Nobody gets a vote on whether OpenAI or Anthropic ships a new model. People do get a vote on zoning and substations. A recent NBC News poll found 69% oppose a data center near them, 70% are more worried than excited about AI overall, and 44% don’t trust either political party to handle it. Whatever this issue is, it isn’t partisan.
Sponsored. Journalists, PR pros and communicators: the fall cohort of AI for Media starts October 13, six live Tuesday sessions with Pete Pachal plus two 1:1 coaching calls. Code AISEARCH500 takes $500 off the $1,500 price for anyone who found the course through AI search, a bigger discount than is offered anywhere else.
The tools genuinely work now, and that’s the problem
Here’s the part that makes this messy: the labs aren’t wrong that the tools are better than ever. I’ve automated most of my own podcast production and promotion with AI in the last few weeks alone. Instead of hand-editing in Premiere, Claude now drives the recording service’s own cloud editor, spins up a virtual machine to finish the job, schedules the release, pushes it to social, and keeps guests in the loop, the entire chain of “lever pulling” that used to eat an afternoon.
That’s not just convenient. It’s money I’m not paying someone else to do the work. Which raises an awkward question: is disclosure even the right instinct here, when the whole use case is automation? Nobody discloses which audio editor they use.
That’s the bind newsroom and comms leaders are actually in. AI can now automate and enhance entire workflows at scale, but most of that adoption still assumes an audience that doesn’t have strong feelings about it. That audience is gone.
Two studies published this year in Digital Journalism get at what readers actually do with this information. Jessica Zier and Nicholas Diakopoulos interviewed readers about AI disclosure and found people say they want the label, then treat it as a red flag the moment they see one. (“I probably need to fact-check this,” one participant said.) A separate team led by Sebastián Valenzuela ran a conjoint experiment in Chile, asking people to choose between outlets with different AI policies across seven dimensions. Human oversight was the single biggest driver of credibility. Readers had no strong opinion about AI on routine tasks, but they trusted outlets that automated both routine and nuanced work less than outlets that banned AI writing outright.
Put those findings together and you get a policy that’s hard to write down cleanly, but a philosophy that isn’t: readers want someone accountable, not a badge. I made this case earlier this year with respect to bylines on AI-assisted articles, and it’s really the backbone of a broader point about naming who stands behind a piece of work.
My podcast is a clean example of that. A messier one, that lands on the same lesson: Cleveland.com reporter Kaitlin Durbin’s byline appeared on an Express Desk story she says she never wrote or reviewed, while she was on her honeymoon. Editor Chris Quinn called it a miscommunication and fixed it within hours. The easy takeaway is “AI bad,” but that skips over the actual context. Cleveland.com built Express Desk so reporters could spend less time typing and more time reporting, and Quinn has said publicly that he sees AI as the future of newsrooms, not the end of them. The failure wasn’t that AI touched the story. It’s that the name attached to it belonged to someone who hadn’t touched it and couldn’t answer for it.
That’s the whole ballgame, and it’s worth writing down as doctrine while the backlash is loud and the tools are genuinely capable. The skepticism was never really about the mechanics of how a newsroom adopts AI. It’s about who’s answerable, and whether a reader can actually find them.
Accountability beats disclosure every time
None of this makes the backlash disappear, and it’s worth newsroom leaders retiring the idea that the right policy will make it go away. What’s actually changed is the cost of adoption, and most newsrooms haven’t rerun the numbers. Eighteen months ago, automating a workflow was purely an efficiency call. Now it’s also a trust call, made in front of an audience that’s already decided what it thinks and is watching for tells. Accountability won’t insulate you from that. But it’s the one thing that keeps people arguing with your work instead of writing it off entirely, and whether they realize it or not, that’s the bet every newsroom is placing right now.
A version of this column appears in Fast Company.
