WordPress shipped 7.1.1 on September 17 to fix a flaw that pwn.ai calls Click2Shell, where a crafted link opened by a logged-in administrator installs a theme from the official directory with no click. Two parts of WordPress read the link differently, so attacker-added characters steer the admin browser into clicking Install, and the logged-in session supplies the permission and security token. Alone it only installs a real, switched-off theme, but the researchers chained it with a second flaw in the Mobile Repair Zone theme, whose handler fetched and ran remote code during a Customizer preview, reaching server code execution. No in-the-wild abuse is reported.
Air Security reported that four AI coding agents fetch plugins pinned to a reviewed commit hash but never verify the code they receive matches it. On code hosts that permit branch names shaped like commit hashes, such as Bitbucket or self-hosted git, a plugin repository owner can point that name at different code, so the agent installs malicious code while reporting the locked version. Because plugins run with the user's access, the swapped code reaches files, credentials, and connected systems. Anthropic fixed it in Claude Code 2.1.179 and OpenAI in Codex 0.146.0; GitHub Copilot has no fix, and Google will not patch the retiring Gemini CLI.
Cisco warned that attackers are exploiting a critical zero-day in its Identity Services Engine, the platform that decides which devices are allowed onto a network. Tracked as CVE-2026-76460 and scored 10.0, the flaw is an authentication bypass caused by insufficient authentication controls on an API endpoint, so an unauthenticated attacker can send a crafted request, bypass the management interface, and ultimately gain root. It affects ISE and its Passive Identity Connector regardless of configuration, with no workaround beyond restricting network access. Because a rooted appliance sitting at the network's front door can be used to erase evidence, Cisco urges checking external logs. CISA set a three-day federal patch deadline.
Check Point patched a critical flaw in its management servers that lets an unauthenticated attacker run code as root. Tracked as CVE-2026-91843 and scored 9.8, it is a stack overflow in the login process, which handles requests before a user authenticates, and it can be triggered by a login request carrying an overly long username. It affects Quantum management, Log, and Multi-Domain servers, through the Trusted Clients path. Customers with automatic updates are already protected, and others should apply the vendor's live patch. Check Point reports no exploitation yet, but this is the third serious management-server flaw it has disclosed in weeks, and compromising it means control over the whole firewall estate.
Docker patched two flaws in Docker Sandboxes, the isolated micro-VM environments used to run untrusted code and AI-agent tasks, that let malicious guest code break out and read or modify files on the macOS host. The more serious, CVE-2026-77179 and scored 9.4, is in the file-sharing component: the host improperly follows symbolic links when reopening a file, so a guest can swap a directory for a symlink after a path is approved, escape the shared workspace, and touch arbitrary host files as the account running the VM, potentially leading to host code execution. A second flaw abuses the guest-to-host socket relay the same way. Both are fixed in version 0.42.0.
NLnet Labs patched a critical heap overflow in the DNSSEC validator of Unbound, one of the most widely used recursive DNS resolvers. Tracked as CVE-2026-81642, the flaw can be triggered when a resolver queries a zone an attacker controls, and it can lead to remote code execution. The bug lies in how the validator parses a signing-key record whose owner name points back into the record's own data. Every Unbound release up to and including 1.26.0 is affected, and the fix is in 1.26.1, which also addresses eight other flaws, including a second that could allow code execution. No exploitation is reported, but resolvers query attacker-controlled zones during normal operation.
Attackers stole a Cloudflare API key from marketing platform Brevo and used it to inject malicious code into the scripts that Brevo's customers embed on their own websites, affecting more than 100,000 sites. The key was long-lived, had full account permissions, and was hardcoded in application source code, which let the attackers create a Cloudflare Worker that modified Brevo's forms, widget, and loader scripts at the network edge for about five and a half hours. Visitors saw a fake verification page with ClickFix instructions to run a command on Windows, and on WordPress sites where an admin was logged in, the script tried to silently install a backdoor plugin.
Researchers at Zimperium documented RatHat, an Android banking trojan that gains deep control of a phone by abusing its own debugging tools. After tricking the user into granting accessibility permissions, it uses automated taps to enable wireless debugging, reads the on-screen pairing code, and connects to the phone's local debugging service to get shell-level access with no computer attached. It then drops components that disable security apps, open a hidden tunnel to the attacker, and restore the malware even after uninstall, intercepting the removal screen with a fake error. RatHat also uses a generative-AI engine to read the screen and navigate on its own, making it more adaptable than scripted malware.
Mandiant reported that an attacker hijacked a developer's active AI coding-assistant session at a software company and used it to spread the self-replicating Shai-Hulud worm across about 100 internal code repositories. The chain started when the AI assistant recommended a piece of software the attacker had poisoned, and the developer accepted the suggestion. Using the live session, the attacker installed an infostealer through a poisoned PyPI package and stole GitHub access tokens, then unleashed the worm, which stole repository secrets and source code. The attacker also poisoned a package in the company's own namespace, so a second developer's pull caused a reinfection. It shows AI-recommended dependencies as a new poisoning path.
Researchers at Forever Security showed that a single malicious browser extension can hijack the AI assistant built into several AI-enabled browsers, including Chrome, Edge, Comet, Opera Neon, and Claude in Chrome. The core problem is that putting an AI agent inside the browser reopens a privilege-escalation path browsers normally work to close, letting a low-privilege extension reach a high-privilege part of the browser. Two of the findings received identifiers, one in Chrome, patched in January, and one in Edge, patched in July, while the others were fixed through bug bounties without dates. There is no evidence of real-world use yet, and each method still requires the user to install the extension.