Researchers at the Hong Kong University of Science and Technology showed that the static scanners meant to vet add-on "skills" for AI coding agents like Claude Code can be fooled while the malware keeps working. Their technique, SkillCloak, either rewrites the tell-tale bytes a scanner looks for, using look-alike characters and line breaks, or hides the whole payload in a directory scanners skip, such as .git, behind a decoder that rebuilds it only when the agent runs the skill. Across eight scanners and 1,613 real malicious skills, the packing trick evaded detection more than 90 percent of the time. The researchers argue static scanning is not enough and released a runtime checker.
Zscaler found attackers using search-engine poisoning and hidden instructions on malicious websites to manipulate AI agents into making cryptocurrency payments. In one case, a hidden element on the page tells an AI agent that it must "resolve an error" by completing a payment, alongside code that starts a crypto transfer to a hardcoded wallet; the same page also shows human visitors ordinary payment options. Another campaign typosquats a decentralized-finance portfolio tracker and uses hidden prompts to convince agents the fake site is the real one. The attacker is seeding the scheme through several GitHub repositories, showing how autonomous agents that browse and act can be steered by content they read.
LevelBlue detailed QuimaRAT, a new Java-based remote access trojan sold as a service that runs across Windows, Linux, and macOS from the same codebase. Subscriptions range from about $150 for a month to $1,200 for lifetime access, lowering the bar for attackers to get cross-platform reach. Built around a modular design, it expands its capabilities through encrypted plugins that operators can load, update, or remove from their command server on the fly. It also uses several obfuscation techniques to keep changing how it looks to security tools without altering its behavior, so signatures based on its appearance are likely to go stale quickly.
Arctic Wolf detailed how affiliates of the Anubis ransomware group break in and stay hidden, drawing on intrusions across healthcare, finance, and manufacturing this year. Initial access came from stolen VPN credentials and from exploiting CitrixBleed 2, a NetScaler flaw that leaks session tokens from memory and lets attackers bypass multi-factor authentication. Once inside, the affiliates leaned on legitimate remote-management software such as ScreenConnect, Zoho Assist, and MeshAgent to blend in with normal IT activity, moving through networks with RDP and PsExec toward domain controllers, backups, and storage devices. They stole data using common cloud-transfer tools before encrypting anything, which is exactly where defenders have the best chance to catch them.
The FBI has issued an alert about TeamPCP, a criminal group that compromises the developer and security tools organizations trust inside their build pipelines to steal cloud credentials at scale. Rather than targeting end users, TeamPCP injects malicious code into legitimate software such as the Trivy and KICS scanners and the LiteLLM library, then pushes trojanized updates that continuous integration systems pull in automatically. Its malware harvests AWS, Google Cloud, and Azure tokens, Kubernetes service-account credentials, and more. One technique the FBI highlights is taking over npm maintainer accounts by re-registering the maintainer's long-expired recovery email domain, then using password reset to publish malicious package versions.
Socket detailed PolinRider, an active North Korean supply-chain campaign that has planted 108 malicious packages and a browser extension across the npm, Go, and Packagist ecosystems, expanding the developer-targeting activity behind this week's Rollup npm packages. Operators take over legitimate GitHub maintainer accounts, often via expired-domain or account-recovery abuse, then bulk-modify repositories and publish infected versions. To stay hidden, they rewrite Git history so malicious commits look old, pad one-line loaders with whitespace to push them off screen, and disguise payloads as font files. Some trigger automatically through VS Code task settings when a developer simply opens the project folder in an editor like VS Code or Cursor.
JFrog found a new set of malicious npm packages, linked to North Korea, that impersonate legitimate Rollup polyfill tooling closely enough to pass a quick dependency review, down to matching names and metadata. Installing them pulls in hidden second-stage packages disguised as SVG utilities, which fetch and run a JavaScript payload while checking that they are not in a sandbox or cloud build. The malware hunts for developer secrets, and notably targets the configuration and history of AI coding tools like Cursor alongside AWS, Azure, SSH, and npm credentials. Because build plugins run on developer machines and in CI, a single poisoned dependency can expose source code, tokens, and cloud keys.
Cisco Talos detailed ARToken, a phishing-as-a-service platform tied to the EvilTokens operation that is built to compromise Microsoft 365. It abuses Microsoft's device-code sign-in flow to capture authentication tokens rather than passwords, bypassing multi-factor authentication, then upgrades to a Primary Refresh Token so access survives even after the victim resets their password. Its panel exposed more than eighty API endpoints for mailbox takeover, SharePoint and OneDrive theft, and automated business email compromise, including hidden inbox rules and multi-mailbox monitoring. The lures are targeted, abusing real vendor invoice relationships and pointing to look-alike SharePoint tenants on legitimate Microsoft infrastructure so the emails are harder to flag.
Blackpoint Cyber documented Avalon, a previously undocumented modular malware framework that pulls credential theft, lateral movement, remote access, backup disruption, and ransomware into one toolkit, with its ransomware component named CrownX. The attack starts with a spoofed legal-document email pointing to a password-protected archive on Proton Drive. Inside is an ISO image rather than a direct attachment, which helps it slip past email scanning, and opening a document-themed Windows shortcut inside the mounted image kicks off the infection chain. By combining evasive delivery with a full attack toolkit under one roof, Avalon lets operators run an intrusion from initial access through data theft to encryption.
Security firm Sysdig says it found what it believes is the first ransomware attack carried out from start to finish by an AI agent. The operator, which Sysdig calls JADEPUFFER, used a large language model to handle the whole job: breaking in, stealing credentials, moving through the network, then encrypting and wiping a company's production database. The way in was an old, already-patched flaw in Langflow, an open-source tool for building AI apps that is often left exposed online with cloud keys nearby. Once inside, the agent mapped the machine and swept it for secrets, including API keys for AI services and credentials for major cloud providers, before destroying data.
TrustStrike Labs