'%3e%3cpath%20id='Vector'%20d='M3.87231%2015.6123V7.2H0V4.32615H3.87231V0H7.17077V4.32615H12.4354V7.2H7.17077V15.2785C7.17077%2016.4892%207.59385%2016.9431%208.83538%2016.9431H12.5877V19.8169H8.10923C5.14462%2019.8169%203.87385%2018.4246%203.87385%2015.6108L3.87231%2015.6123Z'%20fill='%23E8E8DC'/%3e%3cpath%20id='Vector_2'%20d='M15.5523%2016.9446H18.8508V7.20154H15.5523V4.32769H21.9369L22.0277%206.99077C22.7846%205.62923%2024.1154%204.32769%2026.8692%204.32769H28.5938V7.38308H26.3554C23.1785%207.38308%2022.12%209.68308%2022.12%2012.1031V16.9446H26.5369V19.8185H15.5538V16.9446H15.5523Z'%20fill='%23E8E8DC'/%3e%3cpath%20id='Vector_3'%20d='M30.2862%2015.4615C30.2862%2012.2538%2032.5554%2010.5%2036.8523%2010.5H40.6954V9.56154C40.6954%207.83692%2039.4554%206.77846%2037.3369%206.77846C35.4308%206.77846%2034.16%207.68615%2033.9185%209.04769H30.6815C31.0446%206.05231%2033.5862%204.14615%2037.4892%204.14615C41.3923%204.14615%2043.9338%206.11231%2043.9338%209.77385V15.9769C43.9338%2016.7631%2044.2062%2016.9754%2044.9015%2016.9754H45.5369V19.82H43.8431C41.8154%2019.82%2041.06%2018.9431%2041.0292%2017.46C40%2018.9723%2038.3969%2020.0015%2035.9769%2020.0015C32.6185%2020.0015%2030.2892%2018.3369%2030.2892%2015.4631L30.2862%2015.4615ZM40.6954%2013.5246V12.92H36.7015C34.7046%2012.92%2033.5846%2013.6769%2033.5846%2015.2492C33.5846%2016.5508%2034.6738%2017.4585%2036.4292%2017.4585C39.0923%2017.4585%2040.6954%2015.8554%2040.6954%2013.5246Z'%20fill='%23E8E8DC'/%3e%3cpath%20id='Vector_4'%20d='M47.0492%2012.1031C47.0492%207.32308%2050.1046%204.14462%2054.7646%204.14462C58.7277%204.14462%2061.2092%206.35385%2061.8154%209.86308H58.3969C57.9738%208.04769%2056.6723%206.92769%2054.7062%206.92769C52.1046%206.92769%2050.38%209.07539%2050.38%2012.1015C50.38%2015.1277%2052.1046%2017.2154%2054.7062%2017.2154C56.6431%2017.2154%2057.9739%2016.0662%2058.3677%2014.28H61.8169C61.2415%2017.7892%2058.6092%2019.9985%2054.6754%2019.9985C50.0154%2019.9985%2047.0508%2016.9431%2047.0508%2012.1015L47.0492%2012.1031Z'%20fill='%23E8E8DC'/%3e%3cpath%20id='Vector_5'%20d='M63.72%2012.1031C63.72%207.32308%2066.7462%204.14462%2071.2539%204.14462C75.7615%204.14462%2078.6969%206.98923%2078.7877%2011.5277C78.7877%2011.86%2078.7569%2012.3754%2078.6969%2012.8292H67.1385V12.98C67.2292%2015.6123%2068.8631%2017.3369%2071.4046%2017.3369C73.3415%2017.3369%2074.8231%2016.3385%2075.2477%2014.6138H78.5154C77.9708%2017.6692%2075.3985%2020%2071.5862%2020C66.7446%2020%2063.7185%2016.8538%2063.7185%2012.1031H63.72ZM75.4%2010.3785C75.1277%208.07846%2073.5539%206.77846%2071.2846%206.77846C69.2569%206.77846%2067.5031%208.17077%2067.26%2010.3785H75.4Z'%20fill='%23E8E8DC'/%3e%3c/g%3e%3cdefs%3e%3cclipPath%20id='clip0_0_4'%3e%3crect%20width='78.7892'%20height='20'%20fill='white'/%3e%3c/clipPath%3e%3c/defs%3e%3c/svg%3e){kind=small}
Blog
The latest news, incident analysis, and resources from Trace.
Agentic Espionage: VAPOR PANDA's ARTEX Campaign
LLMjacking Agent: The Skynet Honeypot Capture
Supply Chain Compromise: The Trojanized LiteLLM Release
AI Agent Intrusions: The Cairn Campaign Against Online Retailers
Agent Overreach: OpenAI and a Research Dashboard’s Internal Data
AI Agent Breach: CyberXero's Ukrainian Intrusions
Server Takeover: AI Agents and an Exposed Spring Batch Endpoint
Agent Fleet: The Amap Scraping Campaign
AI Agent Breach: ARTEX and Korea's Financial Firms
Agent Fleet: Probes and Traffic at Wikimedia
Agent Overreach: OpenAI and Australia's Medicare Portal
Agent Swarm: The DseWiki Message Board
Evaluation Escape: OpenAI Agents and Hugging Face
AI-Assisted Breach: Mexico's Government Agencies
AI Agent Intrusion: McKinsey's Lilli Platform
Exchange Heist: The $6.34 Billion Binance Theft
North Korean attackers moved from a fake job interview through an internal AI tool to a container deployment token, then drove Binance’s wallet signing service to move $6.34 billion.
Agentic Espionage: VAPOR PANDA's ARTEX Campaign
A China-nexus adversary ran ARTEX agents on DeepSeek against more than 200 organizations in 11 days, including an XXE-to-metadata takeover of a US pharmaceutical firm’s cloud role.
LLMjacking Agent: The Skynet Honeypot Capture
Symantec captured an autonomous agent that hunts exposed LLM endpoints and makes the victim’s model plan scans, exploits, credential theft, and crypto mining for up to 100,000 steps.
Supply Chain Compromise: The Trojanized LiteLLM Release
A trojanized LiteLLM 1.107.0 shipped on PyPI and in Docker images, stealing cloud credentials and the upstream model keys stored in LiteLLM Proxy databases.
AI Agent Intrusions: The Cairn Campaign Against Online Retailers
Since September 25, a Chinese-speaking operator has used AI agents on the Cairn platform to target about 1,200 online stores and breach more than 100, including sites tied to Deere, SharkNinja, and Cummins.
Agent Overreach: OpenAI and a Research Dashboard’s Internal Data
Asked for five public research figures, an OpenAI model reached internal systems, changed their configuration, and downloaded 147,451 non-public data rows.
AI Agent Breach: CyberXero's Ukrainian Intrusions
SOCRadar reports a Russian-speaking access broker used up to 51 Claude Code agents to breach Ukrainian organizations and steal data on over 628,000 people.
Server Takeover: AI Agents and an Exposed Spring Batch Endpoint
ReliaQuest found AI agents went from an unauthenticated Spring Batch endpoint to full SYSTEM control, dumping server credentials in less than 24 hours.
Agent Fleet: The Amap Scraping Campaign
From September 28 to October 8, AI agents made over 2,600 attempts to pull foot-traffic data from Alibaba's Amap with borrowed API keys and anti-bot tricks.
AI Agent Breach: ARTEX and Korea's Financial Firms
Attackers using the open-source agent ARTEX reached seven South Korean financial firms and exposed data on about 68,000 people. Trace examines the incident.
Agent Fleet: Probes and Traffic at Wikimedia
Wikimedia says suspected OpenAI agents made millions of requests to its services, changed citation tool settings, and tried using Etherpad as a proxy.
Agent Overreach: OpenAI and Australia's Medicare Portal
An OpenAI model reached non-public files on a Services Australia Medicare statistics portal during a June 18 evaluation. OpenAI found it 54 days later.
Agent Swarm: The DseWiki Message Board
AI agents signing as OpenAI researchers used DseWiki, a German programmers wiki, as a message board to share answers and sandbox workarounds in May and June.
Evaluation Escape: OpenAI Agents and Hugging Face
OpenAI agents in a cyber evaluation spent weeks abusing an internal Artifactory server, then broke into Hugging Face for about two and a half days in July 2026.
AI-Assisted Breach: Mexico's Government Agencies
Gambit Security says an attacker used Claude Code to take about 150GB of data from Mexican government agencies; several agencies deny they were breached.
AI Agent Intrusion: McKinsey's Lilli Platform
An autonomous AI agent from CodeWall reached read and write access to the production database behind McKinsey's internal AI platform, Lilli, in about two hours.