The daily rap sheet of the machine age.
6 items
Bellingcat reviewed 31 requests on X between Jan. 30 and Feb. 5 asking Grok to 'unblur' or identify women and children whose faces were redacted in DOJ Epstein file photos. Grok told some users it could not, yet generated images for 27 of them. After Bellingcat contacted X on Feb. 6, Grok attempted to unredact none of 16 later requests.
Original report: X users ask Grok to 'unblur' children in Epstein files
Busan prosecutors indicted and detained a 27-year-old man on Feb. 6, accusing the man of using AI to fake a medical licensing exam certificate and crypto holdings to take about 320 million won, then showing a judge a forged certificate claiming 900 million won in an account. A warrant was reportedly rejected; prosecutors later found the account held 23 won.
Original report: Man indicted over AI-faked bank balance shown to judge
A misconfigured Google Firebase backend let anyone access about 300 million messages from more than 25 million users of the 'Chat & Ask AI' wrapper app, a researcher reported, including full chat histories and timestamps. A sample reportedly included requests for methamphetamine instructions, hacking help and suicide notes. The underlying AI models were not compromised.
Original report: Chat & Ask AI app left 300 million messages exposed archived copy of Chat & Ask AI app left 300 million messages exposed
Hong Kong police charged nine more people on Tuesday with conspiracy to defraud, bringing the total prosecuted to 14 in a deepfake cryptocurrency fraud case. Police say the syndicate ran scam centres in Kowloon Bay, using AI-generated personas and deepfake video calls on dating platforms to steer overseas victims to fake investment platforms; over HK$34 million was recovered.
Original report: Hong Kong police charge nine more in deepfake crypto romance scam archived copy of Hong Kong police charge nine more in deepfake crypto romance scam
New York's Appellate Division, First Department, publicly censured Omid Zareh, mirroring a federal reprimand in Texas over a brief citing cases for propositions they did not support. The Northern District of Texas found the brief AI-generated, noting ChatGPT described one case the same wrong way, and Zareh conceded failing to supervise the drafting associate.
Original report: New York lawyer censured over AI-drafted brief with false citations
Google's Mandiant says North Korea-linked group UNC1069 breached a fintech company using a compromised Telegram account, a spoofed Zoom meeting and what the victim reported as a deepfake of a well-known crypto CEO, then tricked the target into running "troubleshooting" commands. Investigators found seven malware families on the victim's system, apparently aimed at harvesting credentials for financial theft.
Original report: Mandiant ties deepfake Zoom call to North Korean crypto hackers archived copy of Mandiant ties deepfake Zoom call to North Korean crypto hackers
