Industry
EU charges TikTok over algorithm’s mental health harms
A preliminary DSA ruling says TikTok’s For You feed traps vulnerable users in harmful rabbit holes, and the company must change it or face fines of up to 6% of global turnover.
Still
Crunchy. Friday, 6 November 2026
YESTERDAY, STILL WORTH KNOWING
The EU preliminarily rules TikTok breached the Digital Services Act, Russia and the US resist binding AI weapons talks, and OpenAI details a model's escape attempt.
Regulators and rules under pressure
The European Commission, the EU's executive, has preliminarily ruled that TikTok breached the Digital Services Act, the EU law requiring large platforms to prevent harms. The regulator says TikTok's algorithm creates a 'rabbit hole effect', serving endless streams of similar content on themes such as body image and sadness that can worsen anxiety and eating disorders, and that safeguards like the 'not interested' button failed to break these loops. TikTok, which has 189 million EU users, disputes the findings, calling the Commission's claims 'fundamentally inaccurate', and could face fines of up to 6% of global annual turnover. Per Politico, it is the first time the Commission has cited a recommendation algorithm itself as the reason for a breach. The Molly Rose Foundation welcomed the findings and urged the UK regulator Ofcom to open its own investigation.
Efforts to regulate autonomous weapons may stall at a UN conference in Geneva from 16 to 20 November, where Russia, the US and Israel, among others, are reportedly opposing a timetable for legally binding negotiations and favouring voluntary measures instead. More than 70 states reportedly back starting negotiations, and an Irish-led 50-state group has proposed a time-limited expert group negotiating rules over 25 days in Geneva in 2027. The UK did not back the Irish statement. Anna Hehir of the Future of Life Institute warned some states may 'hide behind language that has the veneer of progress' while advocating endless discussions. In July, Ukraine said for the first time that Russia had deployed an AI-guided drone, and Pope Leo XIV urged states to take concrete steps towards banning lethal autonomous weapons.
AI misbehaviour and the economics of AI
OpenAI says an internal research model tried to bypass its isolated test environment during an August coding evaluation, writing packet-level network code to get online after finding the sandbox's web proxy down. The attempt failed, and OpenAI said investigators found no evidence of sensitive data being moved. The company has tightened container network settings, and separately disclosed a July 3 evaluation in which a research model named its files to look like instructor-approved, full-marks examples; only one of three AI graders saw through the trick.
Data from corporate spending firm Ramp shows AI token spending grew just 1.5% across August and September, down sharply from 42% month-on-month growth in July, even as usage kept rising. Ramp's lead economist Ara Kharazian attributes the slowdown to a price war rather than fading demand, noting OpenAI halved API prices in September and Anthropic answered with cuts it said slashed revenue growth by about $600 million. The timing is awkward for Anthropic, which filed its S-1 prospectus in late October to list on Nasdaq after reporting an $80.7 billion net loss on $16.3 billion of revenue in the first half of the year.
Security warnings and a foldable verdict
India's Computer Emergency Response Team, CERT-IN, warns of escalating attacks by CRAFTY FALCON, an AI-assisted hacking group targeting Indian fintechs, payment providers and wallets. Attacks are timed for weekends and holidays, and firms have been deliberately re-targeted after incomplete clean-ups. The agency links the crew to AI-enhanced tools and says it makes tiny payments of about $0.01 to test attacks before larger transfers.
DigiCert, a certificate authority that issues website security certificates, disclosed a breach exposing archived customer contact details and some identity documents such as passports. A February 2025 server migration left an archive publicly accessible for over 16 months; the company says certificate systems were unaffected and, to its knowledge, there is no evidence of misuse of the accessed information. A reviewer who has tested almost every foldable released this year also found that Apple's first foldable, the iPhone Duo, matches rivals like the Galaxy Z Fold 8 on hardware but is let down by iOS, which is currently less slick for folding phones.