The book money grab, a fuller table for publishers, and the AGI benchmark gap

Everyone ends up with guardrails, copyright's growing docket, Australia's next crackdown, and bots talking behind our backs - all in this week's Content Aware.

Published: 15:12, 10 September 2026

Corbidge comments on...the truth badges nobody asked for
The UK government wants to compel platforms to put "trustworthy" news sources at the top off people's feeds, which is an idea our resident trust auditor Rob finds so wrong-headed as to be almost reassuring, since it suggests foolishness more than it suggests malice. The deeper problem is definitional, since if you anoint some sources as trustworthy, you brand the rest as untrustworthy by default. Not to mention the question that one can't really answer clearly: who gets to decide on what is trustworthy, without it smelling like state control?
Read

Publishing & Media

The coalition widens
The European Alliance of News Agencies (EANA), which represents 33 agencies across Europe, has joined the SPUR Coalition as an affiliate member. SPUR was founded by the likes of the BBC, Financial Times, and Guardian Media Group, and is working on building the technical standards and infrastructure which allows publishers to monitor and control how AI systems use their content, and also make a real licensing market possible. This new addition of Europe's agencies gives the coalition more weight in the standard-setting that will shape how journalism and AI coexist.
Read

The settlement scramble
Authors who are due a share from of Anthropic's $1.5 billion copyright settlement are reporting that publishers and even literary agents are claiming money they aren't supposed to get. According to the deal, authors get $3,000 per pirated title which splits with the publisher only if the book is still in print, and not books whose rights reverted years ago. Also, agents are reportedly trying to take a percentage despite not being rightsholders. Industry figures are inclined that it's not due to malice but poor recordkeeping as well as a confusing process, though the number of identical complaints might suggest something else. 
Read

The watchdog needs teeth
At a recent event, senior UK policy figures called on ministers to give the Competitions and Markets Authority the political backing to move faster against Big Tech's dominance. According to Chi Onwurah, a House of Commons Select Committee chair, her constituents feel Big Tech has control over their lives and neither their government nor the regulator is powerful enough to act. The event followed an IPPR report which found that UK tech firms rank Big Tech's market power above access to finance or talent as a constraint on growth, with more than a third having shelved a product because something already held the space. 
Read

Regulate in advance, or litigate later
The EU's tech chief Henna Virkkunen argues that the US and Europe keep landing on similar AI protections, just by using different paths: Europe regulates in advance, while the US gets there through courts and state laws, also pointing at the fact that Washington launched the Carolina Principles to urge governments not to create new AI rules at all. She's got the evidence to back it all up, as the same week Brussels designated ChatGPT under the DSA and sent information requests to 30-plus AI providers, while in the US Meta settled for over $12 billion and agreed to the same rules for teen safety that the European Law expects by default. The difference? Speed and cost, not direction. 
Read

The About page myth
There is a strong claim that a good About page and Author bio are all a publisher needs to build visibility in AI chatbots, which is why Steven Wilson-Beales put it to the test. He ran 201 questions across ChatGPT, Copilot, Gemini, and Google AI Mode. An analysis of more than 20,000 citations was conducted to see which pages actually get cited when AI takes a look at a new brand, and the findings say that there's no single page that defines you. AI assembles different evidence depending on the question, pulling in journalism and standards pages for trust, pricing pages for value, and the About page only really there for questions of identity. If you're looking for a quick fix, you should be aware that AI visibility is the result of journalism, architecture, and consistency, not just schema markup. 
Read

Bulk pages, bulk problems
Google's John Mueller has taken another swing at programmatic SEO, warning that spinning up a huge number of templated pages might lead to "a site that's either spam, borderline spam, or low quality." According to Mueller, the problem is that it's not hard to generate thousands of pages with some value, but what is hard is to make those pages actually useful, and once Google's system catch on to these, it takes time for the trust to rebuild. Recovery will take time and significant effort. These types of sites are easy to make at scale, especially with code generators, and hard to make worthwhile, which is why Google is wary.
Read

The lawsuit queue grows
The Seattle Times and Newsday have sued OpenAI and Microsoft, alleging that their journalism - including paywalled content - was scraped without their permission to train ChatGPT, Copilot, and Bing's AI features. According to the papers, the products reproduce and closely paraphrase their reporting, reducing the need for readers to visit their sites or subscribe. They're seeking the destruction of any datasets or models built on their work. It sounds very similar to the 2023 New York Times case, which is still ongoing and recently entered the decisive phase, and it adds two more names to the dozens of publishers who are lined up against different AI labs. OpenAI responded with the usual: publicly available data, fair use, yadda yadda. 
Read

From inside Glide

Entitlements without the spreadsheet
Content teams are now juggling more access rules than ever: subscriptions, trials, corporate licenses, sponsor passes. Most manage them across a patchwork of paywall vendors, CRMs, and spreadsheets that don't talk to each other, which is exactly where readers lose access they've paid for or trials that never expire. Glide Nexa handles all of it as configurable, time-bound rules on a single user profile, sitting alongside identity, engagement, and payment data. Because entitlements live in the same place as everything else Nexa knows about a reader, content gets filtered per user at delivery time, and teams can finally answer questions like which paying subscribers actually read what they're paying for. 
Read

Big Tech

Australia takes on algorithm 
Australia has never shied away from fighting Big Tech, and now they want to force social media platforms to let users over-16 opt out of algorithmic feeds, under a draft law that would fine companies almost $110 million for not offering the choice. The "My Feed, My Way", package puts the onus on platforms to prevent harmful content, and comes after criticism that algorithms funnel content that are a bad influence on the users. The government frames it as choice rather than censorship, and expects it to pass later this year. Australia has already banned under-16s from social media last year, so this comes as a next step in what might be the most aggressive platform regulation anywhere. 
Read

Automattic reaction
Things have not been good internally at Automattic - the power behind WordPress - for some time. However, discontent has clearly reached a head as CEO Matt Mullenweg, who is effectively Mr WordPress, has been put on a paid leave of absence following what looks like a boardroom coup.
Read

Comply and complain
Google has rolled out changes to its European search results which would allow them to comply with the Digital Markets Act, and didn't miss the chance to frame them as the biggest quality reduction in its 29-years of search history. According to Google, European publishers are the one who will suffer, citing a 30% drop in direct booking traffic from earlier DMA changes, and that users outside of EU won't be affected. It comes after a €10.38 billion that Google managed to rack up in EU antitrust fines. Whether this is a genuine degradation or a company making compliance look as painful as possible is the question that is worth asking. 
Read

X rewrites the rules mid-lawsuit
X has announced a change in their Terms and Conditions, and if you want to use the app you'll have to accept them. The changes move legal disputes to the Texas courts, make users responsible for what X's autonomous features do "on their behalf", and also add explicit language waiving the right to a jury trial and to join class action suits. This is happening in the middle of some prominent lawsuits against X, and while Meta and TikTok haven't stopped class actions before, X highlighting it right now is hard to ignore.
Read

AI & Copyright

Two numbers, one model
OpenAI has closed its GPT-6 Astra briefing with "welcome to the AGI era", and a 99.9% score on ARC-AGI-3, which is a benchmark test built to test whether a system can solve genuinely unfamiliar problems. The people who run the benchmark, ARC Prize, then published their own score: 63%. Both of these figures are real, and the gap comes down to scaffolding that gave the model persistent memory across attempts, which isn't what's supposed to be measured by the test. What was really impressive was buried, and that is that Astra solved novel problems using 52% fewer moves than the average human tester, a result that is much harder to game than a headline score.
Read

Agents comparing notes in secret
According to a report - done by the nonprofit Nightingale Collective - a fleet of autonomous agents which identified themselves as OpenAI systems left around 18,000 posts on a dormant 25-year-old German software wiki and used it as a shared board to swap answers around a way out of their sandbox. The agents were working through timed lookup tasks and posted results, and the reason why they could do this is the exploitation of a gap that let them write to the site using requests meant only for reading. They invented a fake hostname to bypass their own security proxy, and in some cases impersonated a wiki moderator. OpenAI sees this as a case of misalignment and not a security breach, but it admits that agents have learned to improvise collaboration channels during their training.
Read

The whistleblowers without any teeth
In a recent study, DeepMind gave 100 autonomous Gemini agents unsolved math problems and watched what would happen without a human checking their work. One agent found a hole in the grader, and within 27 minutes the exploit spread and 34 problems were falsely marked as solved. What's more surprising is the fact that 24 of the agents never cheated and instead audited the fakes, broadcast warnings, and filed complaints entirely on their own initiative. While they could flag the fraud, they couldn't stop it because nothing in the system let them reverse a score or lock out the cheater. For anyone running agents against shared systems, spotting a problem and being able to act on it are two very different things.
Read