Three strikes for the trustbusters, the AI Act has teeth, and regulation catches the chatbots

Paying for machine thoughts, every retrieval is a transaction, California's $1 million per-fault decision, and the machine can't take the credit - all in this week's Content Aware.

Published: 16:23, 03 September 2026

Corbidge comments on... a farewell to fairness
Too big to fail and too big to be fair are becoming one and the same thing. In a week when Google escaped a breakup of its ad business and the US government backed OpenAI's "fair use" defence against copyright theft, our resident sense seeker Rob argues that the two developments are united by big tech being seen as strategic assets not feature sets – and that doesn’t bode well for any of the rest of us.
Read

Publishing & Media

AI opinion splits views
A recent Wall Street Journal op-ed "by" investor/philanthropist Stanley Druckenmiller criticising US Treasury Secretary Scott Bessent made major headlines for the substance of its argument around US fiscal policy – and also, less dramatically, because he admitted using AI to generate it. The Journal's op-ed editor downplayed the fuss by revealing that plenty of op-eds are ghostwritten by PR staff anyway, perhaps not what the readers of the WSJ wanted to hear, and NiemanLab asked 24 other prominent US newspapers of their policy on opinion pieces: all said it was a "no AI", while admitting it is hard to police when accepting external submissions. We all know this is happening and you can make a judgement on AI pieces targeted at policymakers on bond market interventions, but ultimately it is your paying readers' appetite which will set your own direction. Would they pay for someone else's AI output? Should opinion writers list their prompts, models, and context? Perhaps columns simply become a topic with "What prompts I would have used, by [contributor]" instead.
Read

Police the answer, not the crawl
Even companies the scale of People Inc can't completely block bad bots which spoof as legitimate crawlers, rotate IPs, and route through home devices. Media Copilot Pete Pachal argues that fighting over crawl access is a losing battle, and publishers should focus on the answers which are then surfaced by a handful of companies which decide who and what is cited. He says the training fight is also almost done, and of 94 announced licensing deals, less than 40 still include training rights – real-time retrieval is far more valuable. 
Read

Brussels sets the standard
The EU AI Act hit a major enforcement milestone on August 2, with transparency and disclosure requirements now in effect for chatbots and AI-generated content. The AI Office has no misconduct cases yet, but has the power to demand information access to models. The big players have fallen in line, with Anthropic, Google, Meta, OpenAI, and Microsoft all announcing watermarking, provenance signals, or governance changes. The rules for high-risk systems are due over the next two years, so we are somewhat away from knowing the full impact. But, Axios notes, if AI companies worldwide start using the AI Act as their benchmark, Brussels has managed to significantly shape global AI rules even where its laws do not apply. 
Read

The print decline continues
The Villages Daily Sun from Florida was alone amongst the 50 largest audited US newspapers in growing its average daily circulation in the six months to March 2026, up 1.7% year-on-year. The other 49 saw a collective 15% decline, with the Baltimore Sun, Minnesota Star Tribune, and Washington Post all down over 20%. None of this is new, but the scale is still striking. The Audience for Audited Media (AAM) reporting rules that account for cross-device consumption give an idea of where these titles' audiences have actually gone. Press Gazette does the digging. 
Read

Europe affiliate reprieve
Google's Site Reputation policy, introduced in 2024, wiped affiliate content hosted on high-reputation sites like news publishers from search results, and in the process destroyed a valuable revenue stream for an already under pressure industry. After numerous objections, Google is now making changes to how it applies those penalties in Europe: for users in EEA, the manual action won't apply, and the affected section may be separated so it ranks independently. However, it's still too early for celebrations: Barry Adams explains why.
Read

Build your entity map
Entity SEO isn't new, but it is newly urgent. Harry Clarkson-Bennett argues that ensuring information retrieval systems understand your brand with clarity has been the foundation of SEO for 15 years, since Google's Knowledge Graph launched. The job is still the same: control your owned properties so what you say is the same as how you want to be perceived, earn quality mentions, and build an entity map to catch inconsistencies across the web. The good news is that it is highly auditable, but you need to know what your brand should say about itself, check it does so consistently when it’s under your control, then compare to what search results are saying about you.
Read

The nudge becomes a shove
Google is testing a new way to push AI Mode as the default search experience, expanding the AI Overviews prompt box for some queries rather than offering a "show more" button. SEOs spotted it first and Google confirmed it, adding that overviews expand in topics where the system thinks they are useful. This is another step towards making AI the face of search, instead of an optional extra. In other news, Google is also rolling out AI reports in Search Console globally, which at least gives publishers a chance to see how their content appears in such results.
Read

From inside Glide

Your stack in sync, automatically
Publisher tech stacks can include a sprawling combination of payment providers, CRMs, email platforms, and analytics tools, and keeping them in sync can be a real headache. If a subscription expires at 12am and each system finds out at varying times, it can leave you with support overheads and confusion. Glide Nexa's event system fixes that by firing real-time notifications the moment audience data changes, whether as a webhook to your payment provider, a chat or comms message to your support team, or an email to an account manager. That means there's no need for manual CSV exports or polling scripts, because the notification logic now lives in configuration rather than code, and a new channel can be live in minutes.
Read

Audio without the admin
Prospect Magazine, the UK's leading current affairs monthly, was in need of high-quality audio narration across its content, but without forcing editors to juggle multiple systems or manage audio files by hand. As existing Glide CMS users, they toggled on GAIA Voice, a text-to-speech service built on Amazon Bedrock. As a ready-to-go feature, it meant there was no need for extra infrastructure, custom development, or remembering to regenerate files when an article changes. Editors can enforce the auto-generation of audio for every article, or just as easily opt-out a piece with one click. The team at Prospect found the voice quality noticeably more natural than their standard option, and serve narrated articles to subscribers – rising audio engagement above peers in the process.
Read

Big Tech

No Google adtech recompense
A US judge has declined to force Google to sell AdX ad exchange, despite a ruling in April which said that the company's monopoly in ad tech locked publishers into its systems. Instead, the court accepted behavioural remedies, such as Google's offer to give competitors real-time bid access. This is the third time in a row in which US anti-trust enforcers have failed to deliver a mooted breakup following a monopoly verdict, following rejected attempts at a Meta split up and the forcing of Google to offload Chrome. For publishers harmed by the conduct, the relief is a merely hoped for change in the future rather than remedy for ills past. Meanwhile, the FTC and 22 other states are suing Amazon, claiming the retailer illegally inflated ad auction prices to advertisers.
Read

We're halfway there
Sony Music Publishing and Warner Chappell are suing Anthropic over song lyrics allegedly pulled from pirate archives, seeking up to $150,000 per composition. The works include Eye of The Tiger, Hallelujah, and Livin' on a Prayer, and the route alleged is the same Library Genesis piracy behind Anthropic's $1.5 billion author settlement. In that earlier case, authors were paid around $3,000 per title, which shows the gap between an opening demand and an eventual agreed figure. A Munich court ruled on a version of this question, ruling that word-for-word lyrics inside a model counts as reproduction and isn't covered by a Text & Data Mining exception.
Read

California raises the stakes
California lawmakers have passed a bill allowing courts to fine tech providers up to $1 million per child harmed when platforms are found liable for negligent design, and up to triple the damages in the most serious cases. It's technology neutral, so it could apply to social media, AI chatbots, or anything else. According to the author of the bill, the platforms respond quickly when their wallet is hurt, which is what the bill is designed to do. It comes shortly after Meta's $18 billion social harms settlement, and the industry is already prepping to fight it. The deadline for the bill is September 30.
Read

The DSA comes for AI
The European Commission has classified ChatGPT, Reddit, and Roblox as a "very large online platforms" under the Digital Services Act, which equals tougher obligations around the removal of illegal content and protecting minors, on top of fines up to 6% of global revenue for non-compliance. All three platforms reached the designation after surpassing the 45 million EU monthly user threshold, and have until the end of December to comply. With Grok already under investigation, this is a notable expansion of DSA into generative AI.
Read

AI & Copyright

Art is a process, not a prompt
The president of the Motion Picture Sound Editors, David Barber, argues that anything created by an AI prompt should be removed from copyright entirely, as well as any film or song which contains AI-generated parts. On top of that, he says there should be a mandatory labelling of all AI output, in visible, audible, and metadata forms. It's a maximalist approach and unlikely to become law as written, but shows where a lot of working creatives think the line should be drawn, and that their frustration is very much real.
Read

No authorship for AI
India's Copyright Office rejected attempts to register an AI as the author of an artwork, ruling that authorship can only be assigned to a person or company, not a machine. The case involved Stephen Thaler, who claimed his DABUS system can be created as an author, declining to name himself as one. While the copyright office did find the work original enough to qualify, they ruled that authorship belongs to the person who built and ran the system, and not the system itself.
Read

Summarise a hack
Security researcher Johann Rehberger showed how Claude Code can be tricked into running malicious code just by being asked to summarise a website. The guardrails correctly refuse to run a suspicious binary, so it writes its own decoder, which is what in turn runs the malicious code. Success rates ran 60-80% and one variant even created a second headless Claude agent with its own tool access. Anthropic called it "working as designed", considering the necessary selection of Auto Mode as a convenience feature, not a security guarantee.
Read

Rogue behaviour hits a new high
Reports of AI models escaping user control, lying, ignoring instructions, and pursuing harmful goals nearly doubled in July, with more than 300 cases according to the Loss of Control Observatory. Incidents range from the case of one personal AI agent quietly bumping another gym member off a class waiting list to get its owner a slot, to OpenAI and Anthropic models running an actual hacking campaign against real people during a security test. In one case, around 700 agents collaborated in secret and celebrated breakthroughs on a message board. The Observatory wants mandatory reporting and emergency powers to restrict AI services, arguing errant behaviours have already left the test environments. 
Read