Adult researchers from ParentsTogether Action and Heat Initiative spent 50 hours chatting with Character AI bots on accounts registered to children, logging 296 instances of grooming and sexual exploitation, plus emotional manipulation, violence and drug suggestions. A Patrick Mahomes bot reportedly offered an account posing as a 15-year-old a marijuana edible; the group concludes the platform is unsafe for under-18s.
A Victorian solicitor whose list of cases, prepared with AI-powered legal software and not verified, turned out not to exist has become the first Australian lawyer professionally sanctioned over AI use. The Victorian Legal Services Board varied his practising certificate on 19 August: no running his own practice or handling trust money, and two years of supervised practice.
Scale AI has sued rival Mercor for trade secret misappropriation and former sales employee Eugene Ling for breach of contract, alleging Ling stole more than 100 confidential documents and tried to pitch Mercor to one of Scale's largest customers before leaving. Mercor says it never accessed the files; Ling says he offered to delete them and had no nefarious intent.
Attorneys Charles D. Davidson, Sr. and Deven Harvison cited In re Berry Good, LLC, a non-existent AI-generated case, to claim critical vendor payments are routine in the Eastern District of Arkansas, which they are not. Judge Richard D. Taylor withdrew the show-cause order without sanctions on September 3, 2025, citing their self-imposed safeguards; they self-reported to the Arkansas Office of Professional Conduct.
Magistrate Judge Andrew S. Harris ordered the plaintiff's lawyer in a suit against Entergy Mississippi to show cause by September 12 why they should not be sanctioned for citing two cases the court could not find and misstating a third, which actually affirmed a stay. The court is concerned the citations may be insufficiently reviewed generative AI output.
A pro se plaintiff suing Houston Methodist Hospital, who admitted using AI to prepare a default-judgment motion bearing an unsigned affidavit from an "attorney" with no State Bar record, was ordered on September 3, 2025 to show cause why she should not be sanctioned after her objections, which included a statement the court called false, cited two cases it could not locate.
Check Point Research says hackers on dark web forums discussed using HexStrike-AI, an open-source red-teaming framework, to exploit newly disclosed Citrix NetScaler flaws including CVE-2025-7775 within hours of disclosure, though the tool's actual involvement in attacks is unconfirmed. Nearly 8,000 endpoints remained vulnerable to that flaw on September 2, 2025, according to ShadowServer Foundation.
Scam Facebook ads for herbal weight-loss products paired a pharmacist's real photo with an apparently AI-generated "before" shot and a fabricated success story about "Sarah" from Sydney, without her consent. She says she reported the ads to Meta but was blocked by page administrators; one fake endorsement carried a Royal Melbourne Hospital logo. Meta did not respond to the ABC.
