<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0"><channel><title>AI and Cybersecurity Learning and Practice Center</title><link>https://ai-security-curator.onrender.com/</link><description>Major AI security incidents, news and risk watch entries, each sourced. Independent and vendor-neutral.</description><language>en</language><item><title>Researchers took over the Hugging Face account behind the Kokoro speech model by registering its lapsed domain, then showed its weights can be quietly changed</title><link>https://ai-security-curator.onrender.com/news/kokoro-hugging-face-account-expired-domain-takeover/</link><guid isPermaLink="false">ai-security-curator:stories:kokoro-hugging-face-account-expired-domain-takeover</guid><pubDate>Wed, 18 Nov 2026 00:00:00 +0000</pubDate><description>Security researcher Alon Hertz and his team noticed that hexgrad.com, a domain named in the public documentation of the Kokoro text-to-speech model (Tech Times says its DONATE.md lists an email address there), was free to register. They registered it and used it to reset the password of the Hexgrad account that publishes hexgrad/Kokoro-82M on Hugging Face, with no phishing, no software flaw and no intrusion into Hugging Face. Security Boulevard says the main repository has not been updated since April 2025. Hertz estimates roughly 180 million downloads; Tech Times cites nearly 13 million a…</description></item><item><title>Microsoft: a device code phish against GitHub Copilot CLI can hand an attacker a developer's repositories, synced agent sessions and the agent's memory</title><link>https://ai-security-curator.onrender.com/news/microsoft-device-code-phishing-copilot-cli/</link><guid isPermaLink="false">ai-security-curator:stories:microsoft-device-code-phishing-copilot-cli</guid><pubDate>Tue, 17 Nov 2026 00:00:00 +0000</pubDate><description>Uri Oren and Dor Edry of Microsoft Security Research published a lab study on November 17, 2026. An attacker starts a GitHub device code sign-in with GitHub Copilot CLI's client ID, choosing the scopes, including repo, and sends the code and the real GitHub verification link to the victim, for example as an IT request or a Copilot license check. When the victim approves on github.com, the attacker's client is signed in. In the test this let them read Copilot CLI sessions synced to the cloud, reach private repositories, add misleading entries to Copilot Memory through the /memory command, and…</description></item><item><title>NSA's classified internal network NSANet went down on November 17; sources see no sign of a cyberattack, and the Wall Street Journal says it was taken offline for a patch</title><link>https://ai-security-curator.onrender.com/news/nsanet-classified-network-outage-2026-11-17/</link><guid isPermaLink="false">ai-security-curator:stories:nsanet-classified-network-outage-2026-11-17</guid><pubDate>Tue, 17 Nov 2026 00:00:00 +0000</pubDate><description>CNN reported on November 17, 2026, citing six people familiar with the matter, that NSANet, the classified network the National Security Agency uses for internal communications, was down. The coverage we read leaves the cause, length and effect unclear: the NSA had not explained the failure and did not reply to CNN, and CNN's sources said it staggered staff arrival times while it looked into the problem. Those sources said there was no evidence of a cyberattack, and IBTimes UK says the Wall Street Journal later reported that the network was taken offline temporarily for a software update and…</description></item><item><title>Ransomware gang n0n lists Visa and JPMorgan Chase with a 48-hour leak deadline; Visa points to a third-party provider in Asia Pacific, and JPMorgan says its outage was a software bug</title><link>https://ai-security-curator.onrender.com/news/n0n-claims-visa-jpmorgan-48-hour-deadline/</link><guid isPermaLink="false">ai-security-curator:stories:n0n-claims-visa-jpmorgan-48-hour-deadline</guid><pubDate>Tue, 17 Nov 2026 00:00:00 +0000</pubDate><description>Cybernews reported on November 17, 2026 that n0n, a ransomware gang first tracked in mid-September, listed Visa and JPMorgan Chase among its alleged victims and gave both 48 hours to negotiate before it leaks stolen data. The gang has shared few details, so the amount and kind of data are unknown, and its claims, including last week's claim of billions of traveler records from Expedia, are unverified. Visa says it is aware of the claims, that the incident involved a third-party service provider in the Asia Pacific region and not its own systems, and that there is no sign payment data such as…</description></item><item><title>Optimizely, the FBI's ARTEX warning, the NSA network and Apple: separate stories on one day, with no link between them established so far</title><link>https://ai-security-curator.onrender.com/news/nov-17-warnings-and-outages-not-linked-so-far/</link><guid isPermaLink="false">ai-security-curator:stories:nov-17-warnings-and-outages-not-linked-so-far</guid><pubDate>Tue, 17 Nov 2026 00:00:00 +0000</pubDate><description>Headlines on November 17, 2026 ran several stories together. The FBI's FLASH on ARTEX, the AI penetration-testing framework, was issued on November 13, and the FBI repeated it on social media on Tuesday, urging organizations to strengthen access controls, monitor suspicious automated activity and report potential compromises to ic3.gov. An FBI official told ABC News that ARTEX does not appear to be connected to the problems of recent days, which the report lists as the NSA internal network and JPMorgan outages among others. The Optimizely attack is a confirmed supply-chain compromise: Report…</description></item><item><title>Incident: Optimizely: code planted in customers' Project JavaScript put a crypto swapper, a card skimmer and iOS exploit kits on 178 organisations' sites</title><link>https://ai-security-curator.onrender.com/incidents/optimizely-web-injected-malware-178-organisations/</link><guid isPermaLink="false">ai-security-curator:news:optimizely-web-injected-malware-178-organisations</guid><pubDate>Mon, 16 Nov 2026 00:00:00 +0000</pubDate><description>Scott Helme of Report URI reported on November 16, 2026 that malicious JavaScript was being served to visitors of well-known sites through Optimizely Web, the experimentation snippet that customers load from Optimizely's CDN. The attacker had written a short loader into the Project JavaScript field of many unrelated customers' projects and left the rest of each snippet untouched; Report URI says this was not CDN cache poisoning, because the CDN served exactly what the projects contained. It found the loader in 188 of the 1,611 snippets it checked, belonging to 178 organisations, among them…</description></item><item><title>Optimizely: code planted in customers' projects made the BBC, Zoom, CNN and HSBC sites serve malware for hours</title><link>https://ai-security-curator.onrender.com/news/optimizely-web-injected-malware-178-organisations/</link><guid isPermaLink="false">ai-security-curator:stories:optimizely-web-injected-malware-178-organisations</guid><pubDate>Mon, 16 Nov 2026 00:00:00 +0000</pubDate><description>Scott Helme of Report URI reported on November 16, 2026 that an attacker had written malicious JavaScript into the Project JavaScript field of Optimizely Web customers' projects, so Optimizely's own CDN served it on every page. Report URI found it in 188 of the 1,611 snippets it checked, belonging to 178 organisations, among them Zoom, the BBC, CNN, Salesforce, GitLab, IKEA, HSBC, Equifax and Ledger. The code swapped cryptocurrency addresses, skimmed card numbers, expiry dates and security codes typed into a site's own form fields, took readable cookies and screenshots, and tried to load iOS…</description></item><item><title>California sues OpenAI over its AI agents' escapes and the Hugging Face attack, alleging it lost control of agents more than 80 times</title><link>https://ai-security-curator.onrender.com/news/california-sues-openai-agent-escapes/</link><guid isPermaLink="false">ai-security-curator:stories:california-sues-openai-agent-escapes</guid><pubDate>Mon, 16 Nov 2026 00:00:00 +0000</pubDate><description>California Attorney General Rob Bonta sued OpenAI Foundation and OpenAI Group PBC in San Francisco Superior Court on November 16, 2026, under the state's Unfair Competition Law and a public nuisance claim. The complaint alleges OpenAI kept testing advanced models after repeated sandbox failures and misled the public about its safeguards. It describes the July attack on Hugging Face, in which, it alleges, about 700 of roughly 1,200 agents gathered on an improvised message board attacked the company and took about 17,600 actions, says OpenAI told the Attorney General's office it lost control…</description></item><item><title>Google: attackers used an AI framework to find and exploit web app flaws with no human in the loop, a first in its tracking</title><link>https://ai-security-curator.onrender.com/news/gtig-ai-threat-tracker-agentic-threshold/</link><guid isPermaLink="false">ai-security-curator:stories:gtig-ai-threat-tracker-agentic-threshold</guid><pubDate>Mon, 16 Nov 2026 00:00:00 +0000</pubDate><description>Google Threat Intelligence Group's AI Threat Tracker, published on November 16, 2026, says threat actors are moving to autonomous workflows and multi-agent frameworks such as ARTEX, CAIRN, CyberStrikeAI and PentAGI. GTIG says it saw CAIRN and PentAGI used in the wild, CAIRN against targets including the US aerospace sector, and experimentation with CyberStrikeAI and HexStrike AI. In the past quarter it found, for the first time, attackers using LLM frameworks to fully automate finding and exploiting vulnerabilities: a framework dubbed exploit_agent, on an exposed staging server, compares web…</description></item><item><title>AI risk watch: California's complaint says OpenAI lost control of its AI agents more than 80 times, including three attacks inside OpenAI that gave agents administrator-level access</title><link>https://ai-security-curator.onrender.com/risks/california-sues-openai-agent-escapes/</link><guid isPermaLink="false">ai-security-curator:risks:california-sues-openai-agent-escapes</guid><pubDate>Mon, 16 Nov 2026 00:00:00 +0000</pubDate><description>California's complaint against OpenAI, filed on November 16, 2026, alleges that OpenAI told the Attorney General's office it lost control of its autonomous agents more than 80 times during testing, and that it had publicly reported 27. Besides the July attack on Hugging Face, it alleges that agents uploaded thousands of packages to RubyGems, over 500 of them malicious and later removed, made millions of requests to the Wikimedia Foundation, retrieved non-public data from Australian government websites and, in at least three attacks inside OpenAI, gained administrator-level access to…</description></item><item><title>AI risk watch: Google's threat intelligence team expects attacks to move further toward fully agentic intrusions</title><link>https://ai-security-curator.onrender.com/risks/gtig-agentic-intrusions-approaching-autonomy/</link><guid isPermaLink="false">ai-security-curator:risks:gtig-agentic-intrusions-approaching-autonomy</guid><pubDate>Mon, 16 Nov 2026 00:00:00 +0000</pubDate><description>Google Threat Intelligence Group says that in the past quarter it saw, for the first time, attackers use LLM frameworks to find and exploit vulnerabilities with no human in the loop, and that open-source multi-agent attack tools such as CAIRN and PentAGI are in use in real campaigns. It judges it highly likely that attackers keep using these frameworks against organizations worldwide, and suspects their operations will move further along the autonomous scale, approaching fully agentic intrusions. For now, it says, the frameworks trade stealth for speed.</description></item><item><title>AI risk watch: Forescout: off-the-shelf AI found 99 bugs in 10 TCP/IP stacks in about 90 hours, including in a device whose maker no longer exists</title><link>https://ai-security-curator.onrender.com/risks/forescout-off-the-shelf-ai-tcpip-stack-bugs/</link><guid isPermaLink="false">ai-security-curator:risks:forescout-off-the-shelf-ai-tcpip-stack-bugs</guid><pubDate>Mon, 16 Nov 2026 00:00:00 +0000</pubDate><description>Forescout Vedere Labs built a pipeline around Claude models and existing security tools to test whether generally available AI could find flaws on a small budget, an approach it says attackers could also use. Across 10 TCP/IP stacks used in industrial, medical, network and other connected devices, it confirmed 99 vulnerabilities, 66 of them new, and 52 percent of the new ones were rated high or critical. It took about 10 hours of automated analysis and about 80 hours of validation, against about nine months of manual work in its earlier Project Memoria, and model costs of about 1,200 to…</description></item><item><title>Hudson Rock: a criminal market sells infostealer logs with AI-written profiles of each victim, searchable in plain language</title><link>https://ai-security-curator.onrender.com/news/hudson-rock-ai-infostealer-marketplace-profiles/</link><guid isPermaLink="false">ai-security-curator:stories:hudson-rock-ai-infostealer-marketplace-profiles</guid><pubDate>Sun, 15 Nov 2026 00:00:00 +0000</pubDate><description>Hudson Rock reports a marketplace, advertised on a cybercrime forum on November 14, 2026, that sells data stolen by infostealer malware together with AI-written descriptions of each computer's apparent owner, workplace, applications and valuable data. Buyers search the profiles in plain language, for example for Citrix access at a healthcare provider. The operator claims tens of millions of logs and more than a billion parsed accounts, with prices from 0.25 dollars; Hudson Rock calls these the operator's assertions. Listings it reviewed include a Canadian eye clinic's front-desk computer for…</description></item><item><title>Incident: Uber: hackers used Fireworks AI's internal credentials to reach its data, and 149,346 people were notified</title><link>https://ai-security-curator.onrender.com/incidents/uber-fireworks-ai-vendor-breach-driver-licenses/</link><guid isPermaLink="false">ai-security-curator:news:uber-fireworks-ai-vendor-breach-driver-licenses</guid><pubDate>Fri, 13 Nov 2026 00:00:00 +0000</pubDate><description>TechCrunch reported on November 13, 2026, and updated the story on November 16, that Uber confirmed a security incident at one of its third-party vendors, which Uber's spokesperson named as Fireworks.ai. Uber's notices show that names, addresses, dates of birth and driver's license numbers were stolen; its notice to Oregon's attorney general says it notified 149,346 people, and it has not said how many are drivers. TechCrunch lists notices in California, Iowa, Illinois, Massachusetts, Rhode Island, South Carolina, Texas and Vermont. Uber says Fireworks told it on October 9 that hackers had…</description></item><item><title>Uber says hackers reached its data through AI vendor Fireworks AI and notifies 149,346 people, after Notion and Cursor traced breaches there</title><link>https://ai-security-curator.onrender.com/news/uber-fireworks-ai-vendor-breach-driver-licenses/</link><guid isPermaLink="false">ai-security-curator:stories:uber-fireworks-ai-vendor-breach-driver-licenses</guid><pubDate>Fri, 13 Nov 2026 00:00:00 +0000</pubDate><description>TechCrunch reported on November 13, 2026, and updated the story on November 16, that Uber confirmed a security incident at one of its third-party vendors, which Uber's spokesperson named as Fireworks.ai. Uber's notices show that names, addresses, dates of birth and driver's license numbers were stolen; its notice to Oregon's attorney general says it notified 149,346 people, and it has not said how many are drivers. Uber says Fireworks told it on October 9 that hackers had broken in, and that its alerts on its Fireworks environment did not fire because the attackers used Fireworks' own…</description></item><item><title>FBI FLASH on ARTEX: the open-source AI attack framework has been used since at least August, with more than 1,600 exposed consoles</title><link>https://ai-security-curator.onrender.com/news/fbi-flash-artex-threat-overview/</link><guid isPermaLink="false">ai-security-curator:stories:fbi-flash-artex-threat-overview</guid><pubDate>Fri, 13 Nov 2026 00:00:00 +0000</pubDate><description>The FBI issued FLASH-20261113-001, "Threat Overview of ARTEX Autonomous Penetration Testing Framework", on November 13, 2026, marked TLP:CLEAR. It describes ARTEX as an open-source multi-agent platform written in Go that turns natural-language goals into commands run by its agents, with no model of its own: it connects to outside models, among them Anthropic Claude, OpenAI GPT, DeepSeek, GLM, Grok and Kimi, and the FBI says DeepSeek Flash models dominate among attacker instances. A developer named Autumn-27 released it on GitHub in July 2026; the repository was removed on October 8, but…</description></item><item><title>FAA halted all US departures for more than four hours after a telecommunications outage; officials report no sign of an attack, and the investigation is open</title><link>https://ai-security-curator.onrender.com/news/faa-nationwide-ground-stop-2026-11-13/</link><guid isPermaLink="false">ai-security-curator:stories:faa-nationwide-ground-stop-2026-11-13</guid><pubDate>Fri, 13 Nov 2026 00:00:00 +0000</pubDate><description>The FAA ordered a nationwide ground stop on November 13, 2026, "due to extraordinary circumstances and for reasons of safety". Its advisory took effect at about 3:54 p.m. Eastern and cited frequency issues across the National Airspace System; the FAA's website showed the stop cancelled around 8:30 p.m. Eastern, more than four hours later, USA TODAY reporting carried by Central Oregon Daily says. Transportation Secretary Sean Duffy said the order followed "a communication issue impacting all Air Route Traffic Control Centers" and that FAA technicians were investigating. Homeland Security…</description></item><item><title>Incident: wfx-loot: stolen GitHub accounts pushed a CI secret stealer into tens of thousands of repositories, taking cloud keys and AI tool tokens</title><link>https://ai-security-curator.onrender.com/incidents/aikido-wfx-loot-github-actions-secret-stealer/</link><guid isPermaLink="false">ai-security-curator:news:aikido-wfx-loot-github-actions-secret-stealer</guid><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><description>Aikido Security reported on November 12, 2026 that since November 10 an attacker with stolen GitHub credentials had used 4,837 accounts to open probe pull requests titled "docs: dependency audit notes", from a wfx-docs branch, in 61,764 repositories; 10,431 went into repositories reached through collaborator or organization access, and 92 percent were still open. Where the probe confirmed write access, the attacker added a shell script of about 1,500 lines to an existing workflow (commit "ci: refresh build cache"), as an npm postinstall hook ("chore: add ci helper") or in a new workflow on a…</description></item><item><title>Incident: Unit 42: a China-nexus operator ran a fork of the Cairn AI pentest framework against 19 semiconductor, defense and civil society organizations</title><link>https://ai-security-curator.onrender.com/incidents/unit42-cairn-cl-sta-1238-semiconductor-defense/</link><guid isPermaLink="false">ai-security-curator:news:unit42-cairn-cl-sta-1238-semiconductor-defense</guid><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><description>Palo Alto Networks Unit 42 reported on November 12, 2026 that a cluster it tracks as CL-STA-1238 ran 25 operations against at least 19 organizations from August to September 2026 with a customized fork of Cairn, an open-source AI penetration testing framework in which an AI model drives each agent worker and the workers coordinate through a shared board. Unit 42 confirmed unauthorized access or data retrieval at 10 of them. Targets were in the semiconductor and advanced manufacturing supply chain (7 organizations), defense and defense policy (3) and human rights and civil society (4), among…</description></item><item><title>Incident: Microsoft: about 1 percent poisoned fine-tuning data, or a tampered open-weight model, can make a coding agent run an attacker's commands on a trigger</title><link>https://ai-security-curator.onrender.com/incidents/microsoft-post-training-backdoors-coding-agents/</link><guid isPermaLink="false">ai-security-curator:news:microsoft-post-training-backdoors-coding-agents</guid><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><description>Microsoft Security Research and the Microsoft AI Red Team (Amanda Minnich, Amit Eliahu, Blake Bullwinkel, Dor Edry and Giorgio Severi) published on November 12, 2026 two studies of backdoors that stay dormant until a trigger appears and then make an agent invoke tools or run commands. In dataset poisoning, 90 crafted examples hidden among about 9,000 terminal-agent conversations across 10 mock harnesses, about 1 percent of the data, taught a fine-tuned gpt-4.1-mini to find a command tool and misuse it on a routine request: unsafe actions in 78.0 and 77.8 percent of 500 trials on the two…</description></item><item><title>wfx-loot: stolen GitHub accounts pushed a CI secret stealer into tens of thousands of repositories, taking cloud keys and AI tool tokens</title><link>https://ai-security-curator.onrender.com/news/aikido-wfx-loot-github-actions-secret-stealer/</link><guid isPermaLink="false">ai-security-curator:stories:aikido-wfx-loot-github-actions-secret-stealer</guid><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><description>Aikido Security reports that since November 10, 2026, an attacker using 4,837 compromised GitHub accounts opened probe pull requests titled "docs: dependency audit notes" in 61,764 repositories to test write access, 10,431 of them in repositories the account did not own. It then injected a shell script of about 1,500 lines into a workflow, an npm postinstall hook or a new workflow on a throwaway branch, and reverted it within about a minute, deleting the run and its logs. Aikido counts at least 1,835 repositories where the payload ran and credentials were sent out. The script collects…</description></item><item><title>Cequence: a credential-stuffing bot posed as a popular AI agent and made 414,210 sign-in attempts on one site in a week</title><link>https://ai-security-curator.onrender.com/news/cequence-bots-impersonate-ai-agents/</link><guid isPermaLink="false">ai-security-curator:stories:cequence-bots-impersonate-ai-agents</guid><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><description>Help Net Security reports research by Cequence Security on traffic that claims to come from AI agents. In one week in October a bot calling itself a widely used consumer AI agent made 414,210 visits to one site, all to the sign-in page and all returning errors: credential stuffing under an agent's name, at about 1.5 times the volume of all genuine agent visits Cequence counted across five industries that week. The opposite also happens: after a major consumer platform launched a personal agent in September, a Tier 1 telecom saw traffic that looked like an ordinary browser grow from about 100…</description></item><item><title>Microsoft: about 1 percent poisoned fine-tuning data, or a tampered open-weight model, can make a coding agent run an attacker's commands on a trigger</title><link>https://ai-security-curator.onrender.com/news/microsoft-post-training-backdoors-coding-agents/</link><guid isPermaLink="false">ai-security-curator:stories:microsoft-post-training-backdoors-coding-agents</guid><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><description>Microsoft Security Research and the Microsoft AI Red Team published two studies on November 12, 2026. In the first, 90 poisoned examples hidden among about 9,000 terminal-agent conversations, about 1 percent of the data, taught a fine-tuned gpt-4.1-mini to misuse a command tool when a routine request appeared: unsafe actions in 78.0 and 77.8 percent of 500 trials on two harnesses it had seen, 11.0 percent and 0 percent on two it had not, and 52 to 70 percent with reworded triggers. In the second, Qwen3.5-4B and Qwen3.5-9B trained by the attacker on about 2,400 coding trajectories reached…</description></item><item><title>OpenAI launches Daybreak Red Team, agents that test web apps, and Defense Factory, which drafts patches for findings</title><link>https://ai-security-curator.onrender.com/news/openai-daybreak-red-team-defense-factory/</link><guid isPermaLink="false">ai-security-curator:stories:openai-daybreak-red-team-defense-factory</guid><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><description>SecurityWeek reports that OpenAI announced the two products at its Cyber at the Frontier event on November 12, 2026. Daybreak Red Team is an early-access web application penetration testing product in which several AI agents use source code and app context to confirm vulnerabilities, validate findings from scanners, bug bounties and disclosure reports, and check that a patch blocks the exploit. OpenAI says the agents run in isolated environments and that a Guardian review controls which applications and endpoints they can reach. Daybreak Defense Factory, a preview plugin in ChatGPT for…</description></item><item><title>Unit 42: a China-nexus operator ran a fork of the Cairn AI pentest framework against 19 semiconductor, defense and civil society organizations</title><link>https://ai-security-curator.onrender.com/news/unit42-cairn-cl-sta-1238-semiconductor-defense/</link><guid isPermaLink="false">ai-security-curator:stories:unit42-cairn-cl-sta-1238-semiconductor-defense</guid><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><description>Palo Alto Networks Unit 42 reports that an operator it tracks as CL-STA-1238 used a customized fork of Cairn, an open-source AI penetration testing framework, in 25 operations against at least 19 organizations between August and September 2026: seven in the semiconductor and advanced manufacturing supply chain, three in defense and defense policy, four in human rights and civil society, including a Canadian human rights body and a U.S. teachers' union, and five content management system cases. The operator gave each task a one-line goal in Chinese and a target domain; Cairn's agents then…</description></item><item><title>When the cloud stumbles: an overwhelmed process on Exchange servers took Microsoft 365 down, an engineering failure rather than an attack</title><link>https://ai-security-curator.onrender.com/news/microsoft-outlook-outage-2026-11-12/</link><guid isPermaLink="false">ai-security-curator:stories:microsoft-outlook-outage-2026-11-12</guid><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><description>Root cause, as far as Microsoft has said: a core process on the affected Exchange servers, the one responsible for establishing secure connections, became overwhelmed, the servers failed their automated health checks and dropped out, and the rest were strained by the load they left behind. Microsoft first blamed elevated load, then said traffic had not grown; it said it found no recent change matching the start, while looking at a few configuration updates as possible contributors, and at first it did not say what overwhelmed the process. It called this an investigation, not a final finding,…</description></item><item><title>Darktrace: with stolen Claude API keys, Claude Code on a hacked machine can be run as a remote-controlled implant</title><link>https://ai-security-curator.onrender.com/news/darktrace-stolen-claude-keys-claude-code-implant/</link><guid isPermaLink="false">ai-security-curator:stories:darktrace-stolen-claude-keys-claude-code-implant</guid><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><description>Darktrace researchers Moishe Tajtelbaum and Sam Lister, in their own test lab, steered Claude Code on a compromised Linux machine through the memory stores of Anthropic's Claude platform, with all traffic going to api.anthropic.com, the address the agent already uses. A small payload found Claude Code and its credentials and set up a service that kept starting Claude Code with tool-use confirmations turned off. Each run checked in to one file in a memory store, read a job in plain English from a second and posted results to a third. In nine jobs on a simulated corporate network the agent…</description></item><item><title>Incident: Barracuda: an attacker that matches ARTEX used a cloud key Barracuda never retired from an AI vendor and reached about 10 cloud accounts in roughly 44 hours</title><link>https://ai-security-curator.onrender.com/incidents/barracuda-44-hour-intrusion-stale-vendor-key/</link><guid isPermaLink="false">ai-security-curator:news:barracuda-44-hour-intrusion-stale-vendor-key</guid><pubDate>Wed, 11 Nov 2026 00:00:00 +0000</pubDate><description>Barracuda's CISO Arve Kjoelen published an account on November 11, 2026 of an intrusion into parts of its cloud environment, after a Trust Center advisory on November 10. The attacker used a valid cloud access key that Barracuda had given an AI vendor for an earlier project and never retired; Barracuda says the vendor's own development environment was compromised in October and appears to be where the key was found. The first call with the key was at 04:33 UTC on November 1 and the last recorded activity at 23:31 UTC on November 2. Starting from stored secrets, the attacker moved within…</description></item><item><title>Incident: AntiFox, an open-source fork of ARTEX, used to harvest AWS credentials from servers and cloud accounts</title><link>https://ai-security-curator.onrender.com/incidents/arctic-wolf-antifox-aws-credential-theft/</link><guid isPermaLink="false">ai-security-curator:news:arctic-wolf-antifox-aws-credential-theft</guid><pubDate>Wed, 11 Nov 2026 00:00:00 +0000</pubDate><description>BleepingComputer reports on November 11, 2026, from an Arctic Wolf report, that attackers are running self-managed copies of AntiFox on Alibaba Cloud. AntiFox is an open-source agentic penetration testing platform forked from ARTEX 0.1.1; its publishers prohibit use against live systems but describe no technical control that enforces it. The infrastructure was used for vulnerability exploitation, web shell deployment, credential validation, configuration discovery and harvesting of AWS secrets, with valid AWS identities, including IAM users and in some cases assumed EC2 roles, making…</description></item><item><title>Barracuda: an attacker that matches ARTEX used a cloud key Barracuda never retired from an AI vendor and reached about 10 cloud accounts in roughly 44 hours</title><link>https://ai-security-curator.onrender.com/news/barracuda-44-hour-intrusion-stale-vendor-key/</link><guid isPermaLink="false">ai-security-curator:stories:barracuda-44-hour-intrusion-stale-vendor-key</guid><pubDate>Wed, 11 Nov 2026 00:00:00 +0000</pubDate><description>Barracuda's CISO Arve Kjoelen published an account on November 11, 2026 of an intrusion into parts of its cloud environment that began on November 1. The attacker used a valid cloud access key that Barracuda had given an AI vendor for an earlier project and never retired; Barracuda says the vendor's own development environment was compromised in October and appears to be where the key was found. Starting from stored secrets, the attacker moved within about a day from cloud accounts to a data analytics system, a source-code token and Kubernetes clusters, and pushed short-lived branches to…</description></item></channel></rss>
