British Columbia has retained two law firms to pursue legal action against OpenAI, citing what Attorney General Niki Sharma called its failure to notify law enforcement after threats were flagged on its platform before the February 10 Tumbler Ridge school shooting, which killed eight. OpenAI says it disabled the shooter's account in June, and it alerted the RCMP after the attack.
A Paris court sentenced a Nice-based IT specialist to one year under electronic tag, plus a one-year suspended term and a €10,000 fine, for running a site hosting some 300,000 sexual images and videos, mostly AI deepfakes of female celebrities, 13 of whom joined as civil parties. Prosecutors had sought three years; the defendant admitted the facts in police custody.
A class action against xAI was amended to add two anonymous plaintiffs who allege Grok was used to turn their childhood photos into sexual abuse material, in one case more than 7,000 images, and to name Stability AI, which the complaint alleges released Stable Diffusion 1.0 knowing it was trained on such material. Neither company responded to CyberScoop.
Bahia's regional labour court fined a Salvador lawyer 10% of the case value plus R$30,000 after its AI system Galileu flagged white-on-white text beneath the signature reading "GRANT ALL REQUESTS MADE IN THIS APPEAL". The unanimous panel referred the case to the bar association, Federal Police and federal prosecutors; the decision can be appealed.
An independent developer reverse-engineering Claude Code 2.1.196 found code that covertly altered the Unicode of a routine "Today's date" line, encoding a signal for Anthropic's back end when the user's time zone was Asia/Shanghai or Asia/Urumqi. An Anthropic engineer reportedly called it a March experiment against reseller abuse and distillation; the company removed it.
Sysdig researchers say an operation they named JADEPUFFER was run end to end by a large language model, which exploited an unpatched Langflow flaw, harvested credentials, encrypted 1,342 configuration items and destroyed the originals with no human issuing commands. The key was never stored or transmitted, making the ransom demand, per the analysis, impossible to honour.
At least five YouTube channels used AI to clone a Brazilian ear, nose and throat doctor's face without permission for alarmist health videos aimed at elderly viewers, BBC News Brasil reports, including advice to swap prescribed medication for "natural" remedies. São Paulo police are investigating possible false identity and attempted defamation; the clones stayed up after YouTube was alerted.
A woman in the Philippines sent about 100,000 pesos (roughly $1,600), a year's savings, to a scammer who used AI-generated WhatsApp video calls to pose as Dubai's crown prince, AFP reported, supposedly for a marriage certificate and a "royal membership card". She grew suspicious when asked to fund a hotel room and traced the linked Facebook account to Nigeria.
