Speed has always been oxygen in the news business, and the 2010s gave newsrooms an extra reason to breathe deeply. When search and social were the main pipes to readers, the pressure to publish first was constant. Especially around major live events like the Oscars or the Super Bowl, the pressure to post fast often meant preparing “shell” stories in advance, with potential headlines and background information already included.

I’ve made this point before: AI has a tough time with breaking news. Because it takes time for facts to be verified and a consensus to emerge about what happened, AI systems—and in particular Google—tend to shy away from summarizing events in the early minutes or hours of a news event. You would think, then, that speed is a diminishing asset in an AI-mediated news environment.

The reality is messier. Some news publishers are pushing in the opposite direction, opting to publish faster, and with more stories, in the wake of breaking news. For its World Cup coverage, USA Today prepared several shell articles around major games, as Digiday reported. Internal AI systems helped accelerate that process, with human editors altering and publishing them as the games developed. USA Today had already tested the approach during the Winter Olympics and got enough of a lift to run the same playbook, at greater scale, at the World Cup.

Getting into the citation pool early

Fast-turn news isn’t the innovation here. The AI layer is. It’s unclear how long it takes for Google to create an AI Overview around a breaking topic. The Digiday piece cites one test in which AI Mode had access to a breaking story’s information within 10 minutes. AI Overviews appear to move more slowly: One SEO consultant said he had seen them appear within about four hours, and sometimes as long as half a day, while acknowledging there isn’t a lot of good data to go on.

Google may need hours to formulate an AI Overview, but USA Today’s results suggest early publication still pays. Being part of the initial set of sources that compose the answer bestows an advantage for ongoing inclusion—as long as the engine treats you as authoritative and the piece maps to the queries readers are actually typing into AI search.

This is why treating shell articles as an ongoing strategy, rather than a one-off, matters. Having multiple stories around the same topic, linking to each other, is a strong signal. It doesn’t hurt that USA Today is a major domain. There’s also a reporting factor at work: USA Today reporters are physically at the games, gathering exclusive quotes, facts, and perspectives in the follow-up. AI sees all of that and notes the pattern as it considers what to include in a summary.

So is there a first-mover advantage? The evidence is mixed. Being early to a story likely factors into inclusion. Muck Rack analyzed more than one million links cited by major AI systems and found that the highest citation rate occurred during the first seven days after publication. Recency shapes what gets picked, but the first article to hit publish doesn’t automatically beat the fifth.

The takeaway for AI: early counts more than first. And speed is only one input. Established authority—either on a topic or in the news media broadly—is clearly an advantage. A study from SEO tools company SE Ranking that analyzed 75,550 AI Overviews found that, among recognized news outlets, 10 publications received almost 80% of all mentions. The BBC, The New York Times, and CNN alone accounted for 31%.

The unit of competition has changed

The deeper shift is that the ranked link is no longer the unit newsrooms are competing over. Search rankings still matter, but they are increasingly feeding something else: a cluster of sources that an AI system uses to compose an answer. In that world, ranking is a means. Being one of the sources the answer can’t leave out is the actual goal.

The prize isn’t only the click anymore. It’s presence, citation, and narrative authority, the chance to help set the terms of the story before the reader ever lands on a publisher’s site.

That reshapes the newsroom playbook without discarding it. The job is to prepare for predictable uncertainty: map the outcomes you can foresee, the questions readers are likely to ask, and the context an AI system will need to grasp why the event matters. Before news events, consult with your team and AI on possible outcomes, the stories you’d create, and the search queries that people are most likely to ask. Choose the stories you want to be authoritative on, and use AI to help prepare shells and ensure that all your staff is trained up to know what to do.

The trap to avoid is publishing an empty container with a headline and a promise of updates. The winning article is fast, but not thin. It answers the obvious question, supplies the necessary context, links to relevant background, and shows evidence that someone is actually reporting the story. That means writing for two audiences in a single draft: the human who wants the latest developments, and the machine deciding which sources belong in the answer. Background, links, metadata, original quotes, clear sourcing, and visible updates all become part of the same authority signal.

Reporting is still the moat

Then push that authority beyond the first article, not by spraying the same story everywhere but by reinforcing the reporting where readers and AI systems already go to confirm it. The follow-up analysis can become a short video, a podcast segment, a newsletter item, or a social post, and the goal is consistency, not duplication. AI is a great accelerant, but not a replacement for reporters or reporting.

The metrics also have to catch up. Clicks still matter, but they will undercount the value of this work. Newsrooms need to know whether they’re present in AI answers, whether their reporting is showing up (and how prominently), and whether their original facts and framing are making it into the summary. Traffic share is only half the picture. Share of the answer is the other half.

The tactics are there for publishers with the actual reporting to back them up. Speed still creates the opening. Authority determines who owns the answer—and whether winning it is worth anything.

A version of this column appears in Fast Company.

The post Why authority is the new speed appeared first on The Media Copilot.

The New York Times and a group of other publishers are asking a federal court to sanction OpenAI, accusing the company of withholding or destroying evidence in a high-stakes copyright case over how ChatGPT was trained and used.

In a motion filed Thursday in federal court in Manhattan, the publishers alleged that OpenAI misrepresented its ability to search training datasets and ChatGPT output logs for copyrighted news material. According to Reuters, the publishers said OpenAI told the court it could not search its large language models for their work while allegedly concealing that it had already done so “even before the first News Plaintiff filed suit.”

The motion is the latest escalation in the copyright fight between major news organizations and AI companies. It also moves the dispute deeper into discovery, where the question is not just whether AI companies can use journalism to train models, but whether they can preserve, search and produce the records needed to prove what happened.

The plaintiffs include The Times, the New York Daily News and other media organizations, including Ziff Davis and the Center for Investigative Reporting, according to The Associated Press and Variety. The original New York Times article reported that the publishers are seeking legal sanctions against OpenAI, including monetary penalties and other remedies.

The filing does not ask for sanctions against Microsoft, which is also a defendant in The Times’ broader copyright case, according to The Times’ summary of the motion. Microsoft has invested heavily in OpenAI and integrated OpenAI technology into products including Copilot.

“The evidence is in OpenAI’s training data sets and ChatGPT output logs,” the publishers said in the motion, according to The Times. “But instead of just producing that evidence at the start of the case and focusing on the merits of its fair use defense, OpenAI chose obstruction.”

OpenAI rejected the allegations. “As the Times’ case weakens and they’ve been forced to drop claims against us, they’re persisting with their efforts to invade the privacy of people who have nothing to do with this case, including by making these blatantly false allegations,” OpenAI spokesperson Drew Pusateri told Reuters. “We’ll continue defending our users’ privacy and the long-established principles of fair use.”

The publishers allege that OpenAI deleted billions of relevant ChatGPT conversations or made them unsearchable. They also argue that an OpenAI employee later testified that the company had performed multiple searches for news publishers’ content, contradicting earlier representations about the company’s technical limitations.

A sanctions memorandum posted by Ars Technica says the publishers want the court to bar OpenAI from relying on a disputed 20 million-log ChatGPT sample, find that ChatGPT’s output logs include or would have shown substantial use of the publishers’ copyrighted material, instruct the jury on those findings and award fees and costs tied to the discovery fight.

Those remedies would matter because discovery disputes can shape the trial record. If the court finds OpenAI failed to preserve or produce relevant evidence, the ruling could affect what arguments OpenAI can make later and what conclusions a jury may be allowed to draw from missing or incomplete records.

The Times sued OpenAI and Microsoft in 2023, alleging that millions of Times articles were used without permission to train AI systems that now compete with publishers as sources of information. OpenAI and other AI companies have argued that training models on large bodies of text is protected by fair use, a theory now being tested across lawsuits from authors, artists, music labels and news organizations.

For publishers, the issue goes beyond training data. They argue that AI chatbots and AI search summaries can answer readers’ questions using journalism without sending traffic, licensing revenue or subscribers back to the organizations that reported the information. Media Copilot has been tracking the same pressure point in coverage of Google’s AI accuracy problem and The Times’ warnings about AI companies using journalism without permission.

At the same time, publishers are taking different approaches to the AI economy. Some are suing. Others have signed licensing deals with AI companies. The Associated Press announced a deal with OpenAI in 2023, while other media companies have made agreements with OpenAI, Google, Meta and Amazon.

The sanctions motion could increase pressure on both sides. A ruling against OpenAI would give publishers leverage in court and in licensing talks. A ruling for OpenAI would strengthen the company’s argument that publishers are using discovery to intrude into user privacy and commercially sensitive systems.

Either way, the case shows that AI copyright fights are becoming data-governance fights. The central questions are no longer only what AI systems were trained on. They are whether companies can prove it, search it, preserve it and explain it in court.

The post Publishers ask court to sanction OpenAI in escalating copyright fight appeared first on The Media Copilot.

It’s hardly a revelation to say that Google’s AI Overviews sometimes get things wrong. The Gemini-written summaries at the top of search results have been misfiring on and off since they debuted in mid 2024. It feels like Google will never fully live down the infamous “glue on pizza” moment, and the errors come often enough that they always carry the warning, “AI can make mistakes, so double-check responses.”

Nonetheless, AI Overviews are now the reality for anyone (read: everyone) who uses Google. At some point, publishers have to stop treating each new mistake as a curiosity and start treating the system that produced it as their working environment.

This spring, The New York Times commissioned AI startup Oumi to measure the problem. The ultimate finding: The latest version of AI Overviews was accurate 91% of the time. That looks respectable until you run the math against Google’s billions of daily queries. A single-digit error rate at that scale produces millions of bad summaries every hour.

The Times drove the point home by citing BBC tech reporter Thomas Germain, who ran an experiment. He published a fake blog post crowning himself the world’s best hot dog eating tech journalist. Within a day, AI Overviews were repeating the claim, apparently without checking.

The stunt looks silly because the query was silly. But the underlying mechanism isn’t. Germain succeeded largely because he owned the only page anyone had ever written on that subject. It was an information vacuum. For a well-covered topic, a lone rogue post would barely register.

The lens publishers can’t remove

The hot dog stunt is only one failure mode; it turns out AI answer engines can go wrong in several ways. And the stakes for publishers keep rising: AI Overviews now appear in most searches. An April report from AI-visibility startup QuickSEO put their prevalence at 60.23%, and that was before Google’s May I/O conference tightened the loop between AI Overviews and AI Mode, letting users slide from a summary into a conversational follow up without leaving the results page.

Chatbots aren’t the biggest surface here. Google is. People can opt in to ChatGPT or Claude, but they get served AI Overviews whether they want them or not. That default status is what makes accuracy such a load-bearing question. Publishers can’t set the terms of the lens their work passes through, but they still have skin in the game once it does.

Ubiquity isn’t the same as blind acceptance. Trust in AI answers scales with the stakes of the question. A roast chicken recipe gets less scrutiny than a cancer treatment query, even if the entry point is identical in both cases.

By the time a reader decides to double check an answer, the framing has already landed. The summary supplies the vocabulary, sets up the follow up questions and points to what feels worth investigating next. If a publisher the reader trusts is cited in the summary, confidence rises even when the citation is never clicked. I’ve made the case before that citation is a form of value for publishers, but that value depends on the reporting being accurately represented.

Three ways the machine gets it wrong

To map how AI Overviews fail, I spoke to Isis Blachez, the AI lead at Newsguard who runs the organization’s AI False Claims Monitor. She sorts the failures into three buckets, and each one shows up in the Times study.

1. Weak or irrelevant material rises to the top. This is the glue-on-pizza scenario. That recommendation came from a Reddit post written as a joke (we hope), which made it irrelevant to a serious cooking query. The catch is that the post did answer the question head on, and direct answers rank well in AI discoverability. Journalistic content generally performs better in AI engines when it’s optimized for machines. When it isn’t, or when it’s blocked outright, thinner material can grab an outsize share of the response.
   
   “We do [reliability] ratings of news sites,” explains Blachez. “And we saw that for most of the highly ranked sites, they were blocking a lot of the AI bots, and then most of the low-quality sources were giving full access to AI web crawlers.”
2. The AI finds the right source and misreads it. This is the quietest failure mode and possibly the most consequential. Blachez points to a case where multiple chatbots cited Snopes to confirm a false claim that Iran had attacked a Pakistani flagged oil tanker. The Snopes piece was actually the debunking. The machine flipped it.
   
   “Sometimes, even if it’s citing a credible source, it can be incapable of citing it well or retrieving the information correctly,” Blachez says.
   
   The reporting itself is fine in these cases. The machine is the point of failure. This version of the problem is the one that often features in lawsuits against AI companies.
3. The information pool has been poisoned on purpose. The hot dog story is the innocent version of this. The pro-Kremlin Pravda network is the malicious one. It flooded the web with millions of articles across sites designed to look like news outlets, pushing Russian narratives at industrial scale. Coordinated actors publishing similar sounding claims across many domains can manufacture the appearance of consensus and crowd out honest reporting in retrieval systems.
   
   “So what we’ve observed that worked with Pravda is flooding search results,” says Blachez. “It’s like putting the same information with practically the same language, many domains, many times and just dominating narrative on that specific topic.”

Building the machine readability pass

So the answer layer can go sideways because access is blocked, the material is manipulated, or the content itself invites misreads. The AI operator has an obvious duty to raise the floor on quality. What about the publisher?

A lot of newsroom people have quietly written this problem off as somebody else’s, on the grounds that AI systems are a black box. That framing is understandable and mostly wrong. Publishers can influence all three failure modes. Being in the mix means not being blocked. Discouraging misreads means writing for machine comprehension as well as human. Beating manipulation means publishing your own answers to the queries you want to own.

Blocking crawlers is a legitimate choice. Copyright and the absence of any compensation model are real reasons to shut the door. And when journalism is blocked, Google and every other AI company still owe their users a duty of care with the material they do use. But when journalism is available to the AI, publishers have levers to make sure it’s represented correctly.

Every newsroom already runs an SEO pass on its work. The most effective way to shape what AI Overviews and chatbots surface is to run a machine readability pass alongside it. This isn’t just standard GEO hygiene like matching titles to common queries. It means writing so that the tricky parts of a story remain unambiguous to a machine reader, even when they’re already obvious to a human.

In practice, that means saying the quiet part out loud. A human understands that “alleged” applies to a whole run of paragraphs even when the word only appears once. A machine may not carry the qualifier forward.

A short set of questions to run through the pass:

• Are dates explicitly tied to the correct events?
• Is it clear whether an allegation is being reported, verified or debunked?
• Is the primary conclusion stated plainly rather than left entirely to implication?
• Are corrections and updates obvious?
• Does the article distinguish the original source from later repetition?
• Does the headline create ambiguity that the body later resolves?

As with SEO, editing for machine clarity tends to sharpen the human read too. The trade off is that the pass improves the odds. It does not guarantee anything. The goal isn’t “AI proof” journalism. The goal is to strip out avoidable ambiguity and give accurate reporting a better shot at surviving the answer layer.

Publishers can’t dictate what Google says about their work, and they shouldn’t be expected to patch the flaws in someone else’s product. But as AI settles in as a default filter between journalism and its audience, treating that as a reason to disengage stops being a strategy. Newsrooms can still make the truth easier to find, harder to misread and much harder to replace.

A version of this column appears in Fast Company.

The post AI accuracy is Google’s problem—until it becomes a publisher’s appeared first on The Media Copilot.

A year ago, Cloudflare drew a line in the sand against unbridled AI crawling of the internet. Exactly one year later (again on Canada Day) the company took what it says is the next major step on that journey, introducing new tools for publishers and content creators to not just block bots from crawling their content, but charge them for access.

To me, the most interesting part of this is the new Pay Per Use framework. This builds on the existing Pay Per Crawl system, which charged bots whenever they crawled a page. But that straightforward approach didn’t necessarily capture the value of the crawl—once captured by an AI crawler, a piece of content could be used multiple times, in hundreds or even thousands of answers. On the other hand, something could be crawled and never used at all.

Pay Per Use fixes this by compensating the content owner when their content is actually used in an AI answer. Theoretically, if you published something unique, valuable, and optimized for machines to read, it could end up paying dividends for as long as people ask about it. And knowing that is part of the new system, too—Cloudflare promises analytics for content owners so they know how their content is being used. It’s also going to have a better system of telling bots when content hasn’t been updated so they don’t keep re-crawling the same static page over and over.

The system sounds like a sensible evolution to Pay Per Crawl—at least for inference (i.e. AI search engines). For AI training bots, Pay Per Crawl actually strikes me as the better solution since it’s more “one and done.” And how would you measure the value of an individual piece of content in a training set anyway?

All of this depends on a workable payment system, of course, and Cloudflare shared details on how it’s evolving that part of the framework. The new Monetization Gateway is straightforward: 

• a bot tries to access content
• the gateway responds with the payment needed and how to pay
• the bot deposits the payment and gets a proof of payment
• The bot then re-requests the content with the proof
• the gateway checks it and bestows access.

It’s all nice in theory, but this kind of usage-based pricing becomes a bookkeeping nightmare on the content owner’s side. This is one of the big reasons micropayments never took off in digital publishing—the revenue from a small payment by a single customer was never worth the processing hassle.

Cloudflare says its unique position as a content delivery network helps solve these problems. It’s already tracking and classifying the bots, so it’s easy to add the payment credential to the process. There’s no “account creation” or anything like that—the bot just shows the receipt. And it’s all done on an open protocol, with no checkout pages or separate payment API. Apparently, there are advantages to managing traffic for 20% of the web.

Cloudflare is refreshingly honest that its new Pay Per Use system is an experiment. How this all plays out depends largely on adoption, not just by publishers but also by AI companies and data brokers. Lots of people often say that digital publishing needs its Napster moment—when the music industry transitioned from sketchy Napster downloads to the “legit” option of iTunes. But iTunes downloads were aimed at individuals. Nobody typing a search into Google or Claude is deciding what content to pay for. This is all determined at the company level, and companies will always choose to get the best/most data for the least cost.

And that will ultimately come down to a simple equation: Is it less costly to get the data they want via Cloudflare’s system? If it’s not, it will remain merely an experiment.

The post Cloudflare’s new plan could change how AI pays publishers appeared first on The Media Copilot.

Media consumption recently passed a big milestone: people now turn to social media and video networks like YouTube for news more than any other source. The Reuters Institute’s Digital News Report, now in its 15th year, found 54% of audiences now rely on social and video platforms to get their news, putting them ahead of publisher websites at 51% and TV at 52%.

And the trust side of the ledger keeps getting worse. Just 37% of people say they trust most news most of the time, the lowest reading since Reuters started tracking it in 2015. In the United States the figure sinks to 25%. Gallup’s October 2025 poll landed in the same place, with U.S. trust in mass media at 28%, down from 31% the year before and 40% five years ago.

The natural read is that media brands matter less every year, drifting toward irrelevance as audiences scatter into feeds. AI chatbots seem to accelerate the slide. The Reuters report puts news consumption via AI chatbots at 10%, up from 7% a year ago. If brand erosion plus AI summarization is the trajectory, the long-term picture suggests publishers will eventually be reduced to information wholesalers, supplying the raw facts and quotes that someone else interprets, packages, and presents back to the reader.

That story has been the dominant framing for about two years now. But the data underneath doesn’t actually back it.

What the click pattern tells you

Most news consumption on social platforms is incidental. Posts and clips arrive between the workout tips and the gadget ads, and the Reuters report identifies a growing slice—now 12% of people and double the 2020 figure—who run into news only while they’re online for something else. That’s not really audience; it’s adjacency.

Behavior inside AI products reads very differently. Among people who click out of an AI answer, 44% do it to verify the news is correct, against 36% on search and 33% on social. Another 43% click to find out more about the source, versus 35% and 34%. Only 51% click for more detail, well below 59% on search and 60% on social.

That’s a behavioral signal worth paying attention to. Inside an interface designed to strip out bylines and erase visual brand cues, audiences are reaching back through the answer to get to the publisher who supplied it. The dominant reason isn’t curiosity. It’s verification. Readers don’t fully trust the summary, so they reach for the name they recognize to check it.

That breaks the simple “trust in news is collapsing” story. The aggregate trend is real, we don’t live in the aggregate. People can hold low trust in “the media” while continuing to rely on the specific publications they’ve read for years. The Reuters data confirms it: Overall trust fell in 29 of the 48 markets surveyed, but trust in the most widely used individual brands held its ground, with several major names sitting above the broader decline. Behavior and stated preference point at the same answer. Audiences are funneling toward names they already know.

The brand still matters. Arguably more than at any point in the last decade, because the brand is the only fixed object as the surrounding interface keeps changing.

Trust converts but not on impact

We should be realistic about the size of the audience that gets news via AI—it’s still only 10%, and just 1% call AI their main news source. But the slice is growing faster than any other channel, and it skews toward the most engaged readers. Among the biggest news lovers, 18% already use AI for news. That is the cohort every publication has been trying to win for the last decade.

The catch is that trust is not directly convertible. A reader who treats your name as a stamp of credibility inside a chatbot summary may never click. A reader who does click to verify a fact on your site likely arrives, scans, and bounces. Brand reliance at the moment of consumption often produces no measurable lift.

The conversion, however, can happen somewhere else. The reader who keeps reaching for your name to check the machine is the reader who eventually subscribes, who shares your work to a contact, who recommends the publication when a friend asks where they get their information. Reuters found that 46% of paying news consumers now cite values-based reasons for paying, rather than the specific content they’re buying. Those reasons accrue. The brand-reliance behavior happening inside AI interfaces is the leading indicator of the durable reader relationship that eventually shows up in revenue.

The practitioner work for the next 18 months is operational. To make the most of AI audiences, publishers need to build instrumentation that captures the moments when readers reach for the brand, even when the click numbers look thin. Build persuasion strategy that converts those signals into something countable.

Stop playing defense

The headline finding of the Reuters report implies a strategy for media: get on more surfaces, get on them harder, push more short-form video, and lean into the platforms that audiences actually use. Most publishers are following that script. On AI, the script has been the opposite, with many media sites blocking crawlers completely.

All of that is defense. While defense is important, if it’s your entire strategy, you will lose. The offensive posture is to fight to be the default name in your lane, the publication readers reach for when they doubt whatever the surface is showing them.

That means using social, but treating it as funnel rather than destination. Casual readers get a taste; the strategy is to convert a fraction of them into a brand relationship that survives outside the platform. It means blocking crawlers that take without permission, but pairing the block with clean, machine-readable paths for partners and licensees. It means producing the clip, but anchoring the clip to deep, comprehensive coverage that earns the reader’s return visit and, eventually, their subscription.

The creator economy points the same direction. About 27% of people now get news from creators who explicitly focus on news, and 46% from creators of any kind. Those creators score better than legacy outlets on relatability and entertainment value. They also rate lower on trust and impartiality. And the audience that watches them consumes more traditional media than the average reader, not less. Only 3% rely on creators alone. Creators introduce audiences to topics. The brands pick up the verification.

The fragmentation story is real. Audiences are scattering across more surfaces, taking news in smaller pieces, and getting more of it from formats that didn’t exist a decade ago. But the behavior underneath that fragmentation runs the other way. The more the news gets sliced up, the harder readers lean on a name they trust to tell them what’s actually true. Audiences take their news in smaller bites now, but the chef still matters.

A version of this column appears in Fast Company.

The post The news brand is the only thing AI users still click for appeared first on The Media Copilot.

The AI industry pumps out so much hype that you’d be forgiven for simply shrugging at the recent release and sudden withdrawal of Anthropic’s Fable 5 model. Set against the Elon Musk vs. Sam Altman trial and Anthropic locking antlers with the Department of War, the Fable episode could read like just another week in AI.

This one is worth paying attention to. This is really the first time the government has stepped in to regulate a specific model release on the grounds that its capabilities could pose a national security risk in the wrong hands. Whatever happens next, the line has been drawn: a frontier model in general release can be taken off the board because Washington decides it’s too dangerous to leave widely available.

For anyone building AI into their daily work, that shifts the calculation in a real way. The intelligence available to you isn’t only a function of price anymore. It’s also a function of policy, geography, the terms you’re willing to accept on your data, and whether the vendor or the government leaves the model running at all.

The story behind the freeze

For readers who don’t track model releases closely, here’s the short version. Fable 5 is the first generally available model in what the company is calling its “Mythos-class” models, a tier above Opus that Anthropic says has crossed a meaningful risk threshold in cybersecurity and biology. Fable 5 is the consumer-safe version, built on the same underlying Mythos 5 model but wrapped in extra guardrails designed to block or downgrade certain cyber, biology, chemistry, and model-development queries. It also jumps Anthropic’s core model number, signaling a generational step forward from Opus 4.8, Sonnet 4.6, and Haiku 4.5.

Then, on June 12, three days after launch, the government ordered Anthropic to block Fable 5 and Mythos 5 from every foreign national, including foreign-national employees working inside the United States. Anthropic said it could not reliably enforce that distinction and disabled both models globally. The trigger, by most accounts, was a suspected jailbreak that punched through Fable’s cybersecurity guardrails. Anthropic disputed the severity of the finding, saying the demonstration uncovered only minor, previously known vulnerabilities that other public models could also identify.

That fight is still going on. Cybersecurity leaders have urged the government to reverse the order, arguing that defenders need access to the same capabilities and that comparable tools are already available from American and Chinese competitors. Anthropic is working to get Fable back online, and rival labs will almost certainly ship something comparable in short order (some are already claiming to have done so).

The specific dispute may resolve in days or weeks. The precedent will outlast it. A model can be released, integrated into workflows, and then disappear because a government draws a line around who may use it. For anyone building around a single model or vendor (and “building” might simply be leveraging it in crucial, strategic use cases), availability is now part of the risk calculation.

What people saw before the lights went out

Early users got just enough time with Fable 5 to confirm Anthropic’s claims about it. Despite controversies over how Anthropic chose to limit how Fable 5 deals with queries the company deems risky (more on that in a minute), users are seeing the power of the model. Fable 5 is designed for agentic work, meaning it can work autonomously on tasks for a long time, sometimes hours or days, without losing context. The advice that came out of those early sessions was consistent: stop using frontier AI like a fancy autocomplete. The best way to use it, many say, is not to ask it to perform straightforward in-and-out tasks like writing an essay or telling you the best parts of a lengthy report, but to give it broader goals about what you’re trying to achieve, let it build the plan, then execute, however long it takes.

That window was short, but it counted. It showed this level of intelligence is no longer a slide in a research deck. The model was pulled back, but the capability threshold remains crossed.

A big part of what makes Fable work is that it grades its own homework. If you’re a regular user of Anthropic’s models, you’ll notice there’s no “Thinking” mode for Fable 5. That’s because adaptive thinking is always on: The model decides when and how much to reason on every request, and at higher effort levels it can reflect on and validate its own work. Tasks turn into loops. As it works to achieve the goal, it can try things, evaluate the results, change course as needed, and try again. And it can do so autonomously.

For media and marketing teams, the practical shift is in scope. Instead of, say, assigning it to design a specific email campaign, or help format your newsletter, you can zoom out and tell Fable 5 to conceive and build an entire marketing strategy around your newsletter. That might involve reformatting your templates, building new landing pages, adjusting the publishing schedule, building a social campaign, and more. Theoretically, with the right access, it could then build all of that for you. Your job is to grade the output. Over time, less of that grading happens mid-process and more of it happens at the end.

That’s the promise anyway. The danger is that organizations may begin designing around that promise before access, cost, and governance are stable enough to support it.

Fable 5 is the first model that puts real agency on the table. Right now, working with agents, while powerful, involves a lot of management: ensuring the plan the agent builds is correct, clearing up barriers that it encounters as it performs the task, and then guiding it to the best output, usually through multiple iterations on the task itself. In theory, a model strong enough to evaluate its own intermediate work shouldn’t need that hand-holding.

That gap between theory and practice is the real story of the freeze. For a few days, users could test a different relationship with AI; then the capability vanished. We crossed the threshold in the lab and lost it in the market on the same week.

The three walls between you and frontier intelligence

Fable 5 and the models that will follow it stand to change how we work with AI, and arguably how we work, full stop. However, using Fable 5 to its full potential was never just a matter of selecting it in your model picker or calling the API and letting it cook. The pullback put a sharper point on a problem that was already there: the most capable models are also the hardest to actually deploy. I see three walls in the way, with a fourth that just got built.

1. Access and context. For an organization to use Fable 5 to its full potential, it would require a large amount of access to the right context (the org’s information and data). Here, Fable’s strength tripped over itself. Because Anthropic fears the model could be misused, it requires prompts and outputs from Mythos-class models to be retained for at least 30 days for safety monitoring, including in enterprise environments that would otherwise use zero data retention. Anthropic says the data will not be used to train models and that, on some third-party platforms, it remains inside the customer’s cloud environment. But companies cannot use Fable 5 under a true zero-retention arrangement.
   
   That retention requirement, plus the restricted categories where Fable 5 quietly throttles down to Opus 4.8, has set off real friction with enterprise buyers. Many companies will be reluctant to cede control over how their own data is retained and reviewed. Microsoft reportedly limited employee access while its legal teams assessed the implications for confidential and customer data.
   
   And on top of all that sits the new wall. Even if a company accepts the privacy terms, secures the integrations, and builds the right internal controls, the model can still disappear because of a government order or vendor decision. Serious agentic systems will need fallback models, portability across vendors, and a plan for what happens when the most capable model is suddenly unavailable.
2. Compute. Fable 5 is not cheap. Anthropic priced it at $10 per million input tokens and $50 per million output tokens, twice the price of Opus 4.8. I’ve written before about how the agent era is squeezing compute budgets at every layer, and with AI hardening into a political wedge issue, expect compute pressure to stay tight for months and probably years.
   
   The premium price doesn’t automatically kill the math. Some early users argued that it could solve hard tasks in fewer turns than weaker models, potentially lowering the total cost of completing the work. Still, that argument only holds if the work was worth doing with a frontier model in the first place.
   
   If Fable 5 and its peers are going to act as the brains at the top of a company’s AI stack, the deployment question is going to need actual rigor. Organizations will need to be very selective of how to deploy it: which tasks to assign to it, who should have access, and what guidelines, rules, and restrictions there need to be on usage.
   
   And there’s an awkward irony in talking about allocation right now. Intelligence can be technically achievable and commercially valuable while still being unavailable.
3. Task imagination. I became aware of the term “task imagination” through the AI Daily Brief podcast, which references a video by the AI strategist Nate B. Jones. In his take on the Fable 5 release, Jones makes the simple observation that not many knowledge workers think about their work in terms of tasks that might take days to do. It requires a certain level of strategic thinking that may not actually apply to many roles. Put bluntly: a model can run for two days, but most workers have never been asked to define a goal worth two days of machine effort.
   
   For media practitioners, that’s the part worth sitting with. An editor might call on the model to develop more granular editorial guidelines and style guides based on different article types (news, features, evergreen explainers, etc.). Reporters might build investigative agents that don’t just surface data in document troves, but develop research plans based on leads and then execute on them by mining remote databases, filing FOIA requests, and other complex touchpoints that typically require human involvement.
   
   The catch is that most jobs aren’t scoped that way. Many jobs have narrow definitions of what the work is, and there’s little motivation to go beyond that. A model that can do days of work isn’t very useful if the work it’s given is still measured in minutes. That puts pressure on workers to imagine more ambitious tasks or risk being left behind.

The paradox of a pause

The Fable 5 pause comes wrapped in a paradox. A pause gives organizations time to build the governance, data practices, and strategic habits needed to use this level of intelligence responsibly. The trouble is, task imagination only develops with hands on the model. Without access, people cannot discover which long-running assignments are worth the money, where agents fail, or how their own roles could expand around them. The pause buys time while taking away the main way to use that time well.

Step back, and a clearer picture forms. A future where we’re working alongside agents will encounter serious barriers beyond just capability (and political freak-outs over that capability). We restrict access to context so neither the tool nor its creators knows too much. We limit how much we spend on models because we’re unsure of the return we’ll get. And many of us throttle our ambition with AI since our jobs simply don’t have a rich enough canvas for a model like Fable 5 to fill in.

A fourth restriction now sits on top of those three: the model itself may simply not be available.

For media leaders trying to make the ROI case, that’s a problem. The strongest demonstrations depend on giving capable models real work, real context, and enough time to execute. When the most capable models are pricey, hemmed in, or suddenly absent, teams drift to safer pilots that are easier to approve and unlikely to move the underlying economics.

None of these walls fall just because someone ships a smarter model. While advancements in security, infrastructure, and work redefinition will help us get past them, those are inherently slower than the rapid advancement of AI.

We pushed past one threshold and walked straight into several walls. I suspect the story of Fable 5 will be looked back on not primarily as a step up in power, but as the moment where the implications of that power pushed the limits of the systems meant to use it. Agentic AI is clearly where this is going. The systems around it need a beat to catch up.

The pause is useful, but it isn’t free. Experimentation is how organizations learn what this intelligence is actually for. For now, AI leaders are about to discover that running frontier AI at full strength is harder than proving the strength exists. The pure experimentation phase is over. The reality check phase has started, and access, cost, control, and utility now matter every bit as much as raw intelligence does.

A version of this column appears in Fast Company.

The post The Fable 5 pullback turns AI availability into a planning problem appeared first on The Media Copilot.

The European Commission has published a voluntary Code of Practice on Transparency of AI-Generated Content, giving AI providers and deployers a concrete path to compliance with the AI Act’s labeling requirements—and a clear reason to sign up.

The code, released June 10, 2026, covers two broad categories of obligations. Section 1 targets providers of generative AI systems, requiring them to mark outputs—audio, image, video, and text—in machine-readable formats and ensure their detection as artificially generated or manipulated. The technical solutions must be effective, interoperable, and reliable “as far as technically feasible,” factoring in content type, implementation costs, and the state of the art. Section 2 targets deployers, requiring them to label deepfakes (audio, image, or video that falsely appears authentic) and disclose AI-generated or manipulated text publications on matters of public interest.

The Commission also released a set of standard icons that deployers can use to label AI-generated content. Nicholas Diakopoulos, a professor at Northwestern University, shared them on LinkedIn:

The code is currently under adequacy assessment by the Commission and the AI Board. Once it clears that review, signatories can rely on its measures to demonstrate compliance with Article 50 of the AI Act, reducing administrative burden and gaining legal predictability across all EU member states. Non-signatories will have to demonstrate adequate compliance individually, assessed case-by-case by national market surveillance authorities.

Signatories also gain access to Signatory Taskforces: working groups set up to share implementation practices and advance marking and detection techniques across the value chain.

The code is described as a “consistent, practical and proportionate” implementation framework, not a replacement for the AI Act or the Commission’s forthcoming guidelines on Article 50’s scope.

The code was developed over three drafting rounds between September 2025 and June 2026, led by an independent chair and vice-chair. Participants included AI system providers, detection developers, industry associations, civil society organizations, academic experts, and organizations with expertise in transparency and very large online platforms. International and European observers also contributed without voting rights. Two dedicated working groups handled the providers and deployers tracks separately.

Key milestones included a first drafting round starting November 5, 2025, a second round in January 2026, a third round in March 2026, and a closing plenary on June 10, 2026—the same day the code was published.

The post EU publishes voluntary code on AI content transparency appeared first on The Media Copilot.

Look, I’m going to be straight with you. The traditional article is powerful. But it’s only one way to present your reporting.

You spend weeks on an investigation. You publish 3,000 sharp words. But what happens to the data behind it? The full timeline? The quotes that didn’t fit?

What if you could turn that investigation into an interactive experience, complete with clickable timelines, hover-activated charts and tagged insights, in about 20 minutes?

With vibe coding, it’s possible.

What is vibe coding?

The term vibe coding came out of developer culture, but it is no longer just for developers. It’s for anyone who wants to tell a story that harnesses the power of coding.

When you vibe code, you’re building an application with the help of AI by focusing on what you want it to do. Rather than coding with HTML, JavaScript or other technical languages, a builder describes the user experience in plain language to a Large Language Model (or LLM).

You might type a prompt like, “Build me an interactive timeline showing [x, y, z] events.”

Here’s what worked for the journalists I’ve seen succeed with vibe coding:

• Step 1: Pick something simple. Don’t try to rebuild your entire investigation. Start with one article, dataset or interview.
• Step 2: Use a basic prompt structure, like: “Build an interactive [website/dashboard/story] that shows [your content] in [style you want]. Focus on [what matters most].”
• Step 3: After you have something simple, iterate 3-5 times. First pass: structure. Second: visual style. Third: functionality. Fourth: polish.
• Step 4: Share your creation with a colleague. Don’t talk, just watch. See if they click around. If they get stuck, that means you built it for yourself, not your audience. Time to iterate again or start over.

Why vibe coding and journalism make sense

When I taught vibe coding through the Google News Initiative AI Lab, I watched journalists with zero coding experience build interactive financial dashboards, data visualizations and branded microsites, all in about 90 minutes.

“This would have taken our dev team a month,” one person told me. “I did it during our session.”

While you can move quickly to an initial application with vibe coding, you still want to get your product or development support staff on board before launching. The real benefit is that vibe coding lets you prototype faster to see if your idea works before needing to commit resources.

This matters because most newsrooms don’t have a developer on speed dial. At the Adirondack Explorer, a small regional outlet covering New York’s Adirondack Park, journalists are building a civic information product that aggregates town meeting recordings, transcripts and minutes across dozens of municipalities. That kind of project would normally require hiring contractors or a dedicated dev team. Instead, their reporters are building it themselves.

When I worked with VTDigger through the Google News Initiative, they automated campaign emails across four audience segments, work that directly generated $40,000 in donations that likely wouldn’t have happened with manual effort alone. Vibe coding turns “we can’t afford to build that” into “let me show you what I made this morning.”

Think of vibe coding as a creative prototyping partner. Get to 80% quickly. Then decide if you need developer support to get to 100%.

Tips for those who are new to vibe coding

I’m repeating this because it’s important: Start simple. Pick one piece of your big investigation. It could be a dense PDF that needs to be more accessible. It could be data sitting in a spreadsheet. It could even be an interview transcript with great quotes that couldn’t all make it into the article.

When describing what I want to the LLM, I get experimental with my prompts. I’ll type things like, “use the most modern UI and UX interactions and animations to make my charts and graphs more interesting and allow me to parse through the data visually.” Or, “build this in the style of a high-end investigative journalism piece meets Wired magazine’s data viz.”

Then I iterate with edits like, “change the color scheme to match our brand,” or, “pull out more quotes from the sources,” or even, “make [x, y, z] section more prominent.”

Three high-level no-no’s when vibe coding

1. Never use it for final production without verification.
2. Don't use it for anything requiring real-time data without a proper backend.
3. Never publish AI-generated content without independent verification. In high-stakes areas like health, legal, financial or public safety, errors can cause real harm.

A Google AI Overview recently told pancreatic cancer patients to avoid high-fat foods, which is the exact opposite of what oncologists recommend and could jeopardize a patient’s ability to tolerate chemotherapy.

AI can generate beautiful visualizations, but it can also confidently present wrong numbers. For anything where errors could harm your readers, verify everything against primary sources.

Vibe coding tools to try

The main platform I use for vibe coding is Lovable.dev. For a simple interactive graphic such as a timeline, searchable transcript or basic data visualization, you can expect to use roughly 3-8 credits to produce a solid prototype.

More complex builds with multiple views, filtering or light database features can take 15-30 credits depending on how much you iterate. The free tier is typically enough to experiment with small projects, while paid plans make sense if you’re building regularly or refining more advanced applications.

Bolt.new is another tool worth knowing. For a simple interactive project, such as a timeline or basic data visualization, you might use roughly 20,000 to 60,000 tokens depending on how much you iterate. More complex builds with custom logic, multiple components or repeated revisions can exceed 100,000 tokens. The free tier is generally sufficient for small experiments, while larger or ongoing projects may require a paid plan.

Bolt tends to give you more control over the code and works well if you want to edit things directly. Lovable is more beginner-friendly with a cleaner interface for non-technical users.

Both tools let you attach content like article text, CSV files or transcripts, describe what you want in plain language, and get a working prototype you can publish immediately.

You might wonder why you need these tools when you already have ChatGPT or Claude. The difference is output.

When you ask ChatGPT to build you a dashboard, it gives you code snippets you’d need to assemble yourself, often requiring a developer to make sense of it. When you ask Lovable the same thing, you get a working application with a live preview, hosting, and a chat interface to iterate on it.

Lovable is actually powered by Claude, but it wraps the AI in a full-stack builder that handles deployment, databases and design. For journalists without coding experience, that’s the difference between “here’s some code” and “here’s how it looks.”

Go build something new

The article format has served us well. It’s not dead, but it is not the only option.

When thousands of people fly to a conference, share incredible insights and then go home, that knowledge evaporates unless it’s transformed into something people can continue to engage with.

Vibe coding lets us do that better than text-only articles can. Not just for conferences, but for city council meetings, investigative data, community journalism and breaking news.

My hope is you’ll take these vibe coding tips and run. You’ll build interactive story formats your newsroom has never seen. You’ll prototype tools that solve real problems. You’ll make journalism more engaging, more accessible and more honest about its data and sources.

Journalists who can write, build, prototype, ship and transform their own work into new formats will define what news looks like in five years. So, go vibe code something. I can’t wait to see what you build.

This post first appeared in News Media Help Desk.

The post Vibe coding for journalists: Build interactive stories without writing a single line of code appeared first on The Media Copilot.

The Centre Daily Times in State College, PA, has voted to unionize after months of pushback against its parent company’s AI tool—a move that, according to The NewsGuild-CWA, makes it the first newsroom in the union to cite AI adoption concerns as a primary reason for organizing.

As Nieman Lab reported, the Centre Daily Times staff voted to join The NewsGuild of Greater Philadelphia last month. All eligible editorial staff signed authorization cards, and McClatchy voluntarily recognized the union. The catalyst, reporters told Nieman Lab, was McClatchy’s Content Scaling Agent (CSA) tool—an AI system that repackages existing articles into short-form summaries for publication or video scripts—and a March internal meeting where Kathy Vetter, McClatchy’s chief of staff for local news, told staff the company would use their bylines on AI-generated content unless union contracts prohibited it.

Josh Moyer, a senior reporter at the Centre Daily Times, took that as a signal. “It was essentially like, if you’re not in a union, your byline gets used; if you are in a union, we’ll follow what the union says,” Moyer told Nieman Lab. “If we want to control what happens to our byline, that’s the company telling us that we need to form a union.”

McClatchy introduced the CSA tool at the paper earlier this year. Reporters initially published at least one CSA-assisted story per week under a generic byline noting AI assistance. But in late February, management changed the policy: AI-generated content would now carry the reporter’s actual name. Reporters objected that it misrepresented their work to readers.

“When our names go on a thing, it says that this article or video is from that person, but that is just not true in this case,” said Trebor Maitin, a service reporter. Maitin was the first reporter at the paper to have his byline changed to reflect AI assistance.

The NewsGuild-CWA’s president, Jon Schleuss, said unionized newsrooms have had more success keeping AI content clearly labeled: “Unionized newsrooms are the ones where McClatchy’s AI slop gets a clear label. In non-union newsrooms, the AI slop may be carrying a real human reporter’s byline.”

Multiple McClatchy publications have seen byline strikes over the CSA tool, and some have taken labor actions over the tool and related workplace issues. For the Centre Daily Times, the union opens the door to formal collective bargaining and the ability to join coordinated actions at sister publications.

“Some of us use AI a lot more and are okay with it,” Maitin said. “But there is an overall understanding that we need to be able to have a say in this, and that unionizing at least gives us a seat at the table.”

The post A newspaper unionized because McClatchy put reporters’ names on AI content appeared first on The Media Copilot.

Reuters and Time are blocking all AI bots by default and only letting approved crawlers through—a whitelist approach that more publishers are adopting as the volume of unauthorized scraping grows.

As Digiday reports, both publishers moved to block AI bots last month, joining People Inc. and The Atlantic, which adopted similar strategies earlier this year and late last year respectively. The goal is simple: content costs money to produce, and AI companies have been taking it without paying.

“We saw that there was an imbalance between the value that publishers like Reuters provide and the value that Reuters receives in kind, and so instead we went from a default allow-all to a default disallow all,” said Josh London, head of Reuters Professional, which oversees the direct-to-consumer and direct-to-professional businesses. Reuters has since signed AI licensing agreements with Microsoft and Meta, according to the report.

The publishers aren’t relying on any single tool. Reuters uses robots.txt files, a method that is voluntary and non-binding, and one that many AI bots simply ignore. The approach is meant to create friction and signal that access requires negotiation. “If you want this, let’s have a conversation and then we can allow you to access,” said Alphonse Hardel, head of agency at Reuters, who leads the content licensing business.

Time allows roughly 70 bots on its site, ranging from AI lab crawlers and social platforms to its own operational systems. The volume of bot traffic has become significant enough that Time sees it as leverage for a future AI visibility product it’s developing for brand clients.

The economics are also shifting. Blocking bots cuts server costs: Hardel said the expense of the bot-blocking vendor can be nearly offset by the reduction in non-human traffic. At People Inc., the shift from a block list to an allow list meant going from blocking roughly 2,100 user agents to over 30,000, said Lindsay Van Kirk, the company’s SVP of innovation, speaking at an IAB Tech Lab event in May.

“Adding two full seconds of latency to the majority of scrapers when you implement a block-all-bots approach is a really good thing, even if they have to go through,” Van Kirk said. “Every scraper who has to pay a home proxy network in order to get access to the content is margin that you are taking out of their business.”

The IAB Tech Lab has published guidance on bot management, and the SPUR Coalition—a publisher group formed earlier this year with major news organizations—announced significant new membership as it works to create technical standards for AI licensing and content protection.

For Reuters, the change hasn’t reduced site traffic. After monitoring bot activity over an extended period, the company had enough data to identify which bots it could block without hurting revenue. The publisher maintains a public robots.txt file that lists approved bots, a benchmark that also supports enforcement discussions, said Phil Andraos, general manager of Reuters Digital.

“It’s not a set it and forget it approach,” London said. “The value of content is something that we ignore at our own peril, especially as AI scales.”

The post Reuters and Time flip the script on AI bots with blocking whitelists appeared first on The Media Copilot.