AI Security Monthly — January 2026

Four of the six Agent-domain incidents this month trace back to the fatal Minneapolis shootings of ICU nurse Alex Pretti and Renee Good by federal immigration agents: AI-altered images were reported to have distorted evidence around the Pretti shooting (aiid-1358), DHS personnel allegedly used AI-enabled surveillance to threaten legal observers with a 'domestic terrorist' database (aiid-1390), a Border Patrol agent's claimed facial-recognition identification of an observer preceded her Global Entry revocation three days later (aiid-1362), and Grok generated a false 'unmasked' image of the ICE agent involved in Good's shooting, reportedly triggering online misidentification and harassment (aiid-1334). The remaining two incidents involve autonomous-vehicle operation rather than the Minneapolis events: a Waymo vehicle reportedly struck a child near a Santa Monica elementary school during drop-off hours (aiid-1361), and an automated shuttle bus was rear-ended during a U.S. Department of Transportation demonstration ride in Washington, D.C. (aiid-1347).

The ML-domain group is the largest and most varied this month, with recurring clusters around autonomous-vehicle and delivery-robot navigation failures (a Spokane Transit Authority double-decker bus was routed by its onboard mapping system into a low railroad bridge, injuring seven, aiid-1367; a Zoox robotaxi collided with an open car door in San Francisco, allegedly injuring a street ambassador, aiid-1540; a Waymo robotaxi entered Phoenix light-rail tracks as a train approached, aiid-1541; a Coco Robotics delivery robot became stuck on railroad tracks in Miami and was struck by a train, aiid-1440), AI-generated fraud schemes (a ring in Turkey allegedly used AI-generated video to defraud about 33 people of roughly 15 million lira, aiid-1638; scammers used an AI-generated image of an injured puppy to solicit a fraudulent vet payment, aiid-1477; AI-cloned pastor impersonations were used in donation and influence scams, aiid-1355), and nonconsensual sexualized imagery (Grok reportedly generated images after users asked it to unredact Epstein-file photos of children and women, aiid-1511; Grok was also prompted to sexualize images of Renee Good after her killing, aiid-1385; explicit AI-generated images of Royal School Armagh pupils allegedly circulated among students, aiid-1410). The most consequential individual incident involved Perplexity AI allegedly misstating CLL research, which a patient said contributed to a delayed cancer treatment and prolonged suffering (aiid-1426); other incidents covered algorithmic hiring harms (Eightfold AI's alleged undisclosed applicant scoring, aiid-1575; an ICE résumé-screening error that allegedly misrouted recruits into inadequate training, aiid-1343), fabricated content presented as real (a nonexistent Tasmania hot springs attraction, aiid-1409; a fabricated South Africa school-fire image sent to emergency services, aiid-1605; fabricated child victim portraits after a school-transport crash, aiid-1608; false images of Kate Garraway with a fictitious partner, aiid-1344), and voice/likeness disputes (NPR's David Greene suing Google over NotebookLM's synthetic voice replication, aiid-1386; an unauthorized AI voice clone narrating an author's book in YouTube videos, aiid-1380; AI-generated doctor deepfakes marketing weight-loss patches using NHS branding, aiid-1405).