Industry
German courts hold Google liable for its AI search answers
Munich and Aachen regional courts treat AI Overviews and AI Mode as Google's own statements, and a quietly settled appeal means no higher court has pushed back yet.
Still
Crunchy.
YESTERDAY, STILL WORTH KNOWING
Sunday's briefs circle a single theme: who answers when AI goes wrong, from German courts holding Google liable to Meta quietly softening its own terms.
German courts make Google answer for its AI answers
Two German regional courts have now ruled that Google is directly responsible for what its AI search features say. Munich Regional Court I held the company liable in May for AI summaries falsely linking two publishers to fraud, and the Aachen Regional Court followed on 24 September, granting an emergency injunction over AI Mode claims that a local neuroradiology professor had assaulted his secretary, an invented scandal borrowed from an unrelated criminal case. The judges reasoned that Google's AI produces independent new statements rather than merely making third-party content findable, so Google answers for them, and noted the company could have caught the errors by checking its output against its own sources. Breaches carry fines of up to 250,000 euros each.
The wrinkle is that the legal test has never actually been tested. Google announced an appeal in Munich, but the case quietly ended in a settlement and no higher court has weighed in. If courts keep treating Google as a publisher while the company settles behind closed doors, the underlying question stays untested, and the reasoning could eventually reach publishers whose texts feed AI answers, potentially unsettling the whole business model of AI search in Germany.
AI accountability gets prodded from other directions
Meta has removed language from its commercial terms that legal experts read as permission to scrape any publisher's website holding a Meta business account for AI training. The company insists the consolidated crawling terms only ever covered ads and business products, but the episode shows how much scrutiny now falls on training-related wording buried in terms of service.
Anthropic's red team, meanwhile, built an AI worm that spread across 15 smart televisions, drawing a little computing power from each. Security experts are split: some call it nifty stunt hacking that basic hygiene would stop, while figures such as former National Cyber Director Chris Inglis argue the plausibility question is already settled and the real question is how such worms would be countered. Anthropic itself says current devices lack the capability, though trends could change that by year's end.
Memory prices squeeze consoles, laptops and data centres
The AI data centre boom keeps pushing component prices up. Insider KeplerL2 claims PlayStation 6 memory and storage alone will cost Sony over $680 per console, and analysts now debate whether next-gen machines will top $1,000, with Xbox facing the same squeeze. The shortage has not slowed the buildout itself: SemiAnalysis finds over 300 local US moratoriums have delayed only three projects, since pauses rarely match the site, permit or timeline, though a Texas state-level permit freeze looms as a bigger threat.
On hardware, The Verge's week with Nvidia's Arm-based RTX Spark laptops found Windows on Arm finally liveable and gaming surprisingly decent through emulation, but at $2,599 to $6,999 these are machines for developers, not everyone. A modder has meanwhile brought Nvidia's full DLSS 5 suite to three generations of AMD Radeon cards, with results close to native but mixed performance reports. And in Washington, Rep. Ritchie Torres's Power To The Players Act would make buying a game mean owning it, requiring offline modes before servers shut down, prompted by NBA 2K25's imminent delisting.