Poetry breaks AI, Sweden unites against Meta, and Google AI's headlines lies

EU pause brings pain, Australia draws a line, and publishers united - all in this week's Content Aware.

Published: 15:02, 04 December 2025

UK's OBR budget leak exposes the real cost of "cheap" publishing tools
Who says technology and publishing is easy to get right? Not the UK's government budget watchdog, the Office of Budget Responsibility, that's for sure. A leak on the organisation's website put embargoed national financial data into the public domain early, and led controversy right to the doors of the country's leaders. An investigation revealed a picture of sensible people doing mostly sensible things - but underscores the old adage of using the right tool for the job, something countless publishers and others in critical publishing roles have learned to their cost over the years.
Read

Zuck in a muck
Swedish media groups are taking Meta to court over scam ads on Facebook and Instagram which impersonate journalists. Swedish media body Utgivarna says the ads exploit media brands and their staff, and cause financial and psychological harm to the faked journalists and scammed users seeing the ads despite repeated efforts from media to get them removed. Meta, recently revealed by its own estimates to have made $16bn in 2024 from the 15 billion dodgy ads per day users see, says that fighting scams is a "top priority".
Read

Versus verses
AI models fooled by rhyme, words concealing tricks each time. In coded lines, a secret spun, Italian boffins show it done. "Adversarial poetry," they dub such plays, unpicking bots in clever ways. A sneak attack with clever pun, could be your hacker's smoking gun!
Read

Legal reality checks 
The legal system is throwing many a wrench into the AI industry's gears, but why do seemingly similar cases take such different turns in rulings? Recent outcomes of suits in the UK and Germany highlighted the subtle differences in direction judges can take in cases, notwithstanding the different jurisdictions involved. Even within the EU, cases have and will go different ways dependant on local interpretations of law.
Read

"Are ya winning?" 
Meanwhile as multiple AI legal actions stack up in courts around the world, Press Gazette AI hawk Charlotte Tobitt has kicked off a rolling monitor rounding up all the twists, turns, and legal drama for you to bookmark. This is in addition to the title's ongoing monitor of who is suing, who is doing deals.
Read

EU pause sparks compliance headaches
EU regulatory whiplash, again. The bloc has paused parts of the EU AI Act to help companies catch-up with the regs, but critics bemoan it means 16 months of uncertainty with no guarantee current plans won't end up being the requirement. What was supposed to be a helping hand to aid industry has turned into a mess, leaving countless businesses and industries unsure whether to push ahead with compliance efforts or stand down and wait.
Read

Screen-time showdown
Australia's social media ban for under 16's kicks in next week (Dec 10), and the message to platforms is obey or face consequences. Social media firms will have to take "reasonable steps" to make sure kids aren't on their sites. The law itself has sparked a mix of admiration and criticism, worry that enforcement will be impossible, and already counter actions in court by teenagers. According to the Australian Communications Minister, this is only the first step in an evolution by Governments to online risks of all kinds. Everyone is taking notes.
Read

Journalists 1, AI 0
AI gets no free pass in the Politico newsroom, as unionised journalists win a landmark arbitration ruling that new AI tools breached contracts. The ruling says new AI tools violated both bargaining rules and journalistic standards, giving Politico 60 days to negotiate or fix.
Read

Discovery phase
Google will tell you misleading headlines is not good for users - but that isn't stopping it from using its own (assumedly cheapest) AI to rewrite yours and fill them with lies. Google AI-generated headlines and summaries are appearing into Discover, with the occasional mangling of facts as bonus. It's being brushed off as a growing pain. But there's more! AI Mode is moving in for a co-starring role on mobile Search too, pushing publishers another step further away from users.
Read

Brothers in arms
The News/Media Alliance has jumped into Thomson Reuters's legal fight with Ross Intelligence, where it is arguing that Ross built an AI assistant by copying Westlaw, Reuters's own legal research tool. The Alliance has warned that if the court green lights this kind of "copy now, innovate later" approach, publishers' burgeoning business of licensing content to AI could get rolled over in the process. Read the full brief here [PDF].
Read

Fake websites, real AI ambitions
Silicon Valley is set on training AI to use websites and carry out actions, and the small matter of being blocked from websites isn't deterring them. The current hot dev ticket in Silicon Valley is building fully working fake versions of sites of major brands and destinations, including Amazon, Gmail, retail and travel sites, and more. They look almost identical - some have even used official logos - with "working" buttons, menus, and such, so AI agents can treat them like school and learn to shop, book flights, and manage tasks. Even though logos are usually different, legal experts believe these are still skating on thin (copyright) ice. The unusual thing is them being made public.
Read