AI Security Monthly — June 2026

June's two Agent-domain incidents both involved automated vehicle systems associated with physical harm or near-harm. An unoccupied Zoox robotaxi reportedly entered a smoke-obscured active fire scene before braking hard, prompting a recall tied to heavy-smoke detection (aiid-1602). Separately, a Tesla Model 3 reportedly left the roadway and struck a home in Katy, Texas, killing resident Martha Avila, with the driver reportedly telling investigators that driver-assistance mode was engaged at the time (aiid-1552). Both cases center on how automated driving systems perceived and responded to real-world hazards.

The sole Identity-domain incident this period involved a purportedly AI-powered platform called KT, which traffickers at Myanmar's Tai Chang scam compound reportedly forced a trafficked worker, Safeer Mohammed Koorimannil, to use to impersonate women and target thousands of people online (aiid-1588). Unlike other impersonation cases in this period, the tooling was deployed inside a forced-labor scam operation rather than by an independent scammer, tying AI-enabled identity impersonation directly to human trafficking.

LLM-domain incidents this period split into two recurring patterns: chatbot output leaking unedited into official work product, and AI-generated content used in financial or political deception. On the leakage side, a staffer in Rep. Anna Paulina Luna's office reportedly pasted Claude transcript residue into an NDAA amendment summary (aiid-1641), Canadian MLA Bill Oliver reportedly read aloud unremoved LLM drafting instructions during a New Brunswick legislative speech (aiid-1607), and a 3M-retained expert reportedly submitted a largely ChatGPT-generated report seeking to show 3M was "0% at fault" in an explosion lawsuit (aiid-1645). On the deception side, a deepfake of Taoiseach Micheál Martin promoted a quantum-AI investment scam (aiid-1549), deepfake ads fabricated a BBC Question Time confrontation between Andrew Bailey and Nigel Farage (aiid-1527), and Reddit ads used deepfake news segments impersonating the BBC, Financial Times, and Guardian to promote purportedly fake AI investment platforms (aiid-1526). OpenAI also reported PRC-linked ChatGPT accounts running covert influence operations targeting U.S. AI policy debates (aiid-1580).

The ML-domain group was the largest this period and was dominated by non-consensual AI-generated imagery used to harass, defraud, or exploit individuals: purportedly AI-generated sexual images of 22 women in Silleda, Galicia (aiid-1614), AI-fabricated sexual content used to blackmail a relative in Egypt (aiid-1560), and a purported AI-generated video allegedly contributing to a Laredo, Texas teenager's death by suicide (aiid-1591). A second cluster involved AI-generated content spreading false claims about real people, including DuckDuckGo Search Assist falsely stating that Donald Trump and JD Vance had died of rabies (aiid-1644), a fabricated Facebook poster declaring Kannada actor Doddanna dead (aiid-1550), fake ads using singer Lee Ji-hye's likeness to endorse products (aiid-1551), and Storm-1516's fabricated narratives targeting a data center under construction in Armenia (aiid-1581). The remaining incidents were unrelated to synthetic media: Cherokee County, Georgia deputies allegedly misused automated license plate reader data for non-law-enforcement purposes (aiid-1595), UDR was alleged to have used RealPage pricing algorithms to set San Diego rents (aiid-1618), and a humanoid robot reportedly kicked a boy during a public demonstration in Xinjiang (aiid-1536).