Cybersecurity Weekly Briefing, 7-13 November

Fortinet patches nearly 100 vulnerabilities, including Fortinet2Shell in FortiClient

Fortinet published 89 security bulletins addressing nearly 100 vulnerabilities, around 17 of which are critical, three weeks after announcing that it had incorporated cutting-edge AI models into its product security testing. FortiMail receives the largest batch of critical fixes, followed by FortiAuthenticator, FortiOS, FortiManager and FortiAnalyzer, with flaws that allow code execution, authentication bypass, account takeover or the execution of commands as a super-administrator.

The most notable is CVE-2026-121529 (CVSSv3 9.6 according to Fortinet), known as Fortinet2Shell: an origin validation error caused by an incorrect CORS configuration in a local FortiClient service for Windows and Mac, which accepts requests from web pages such that simply visiting a malicious site is sufficient to execute code – reliably so if FortiPAM is enabled. It affects versions 7.4.0–7.4.7 and 7.2.0–7.2.14 (Windows and PAM) and 7.4.0–7.4.7 (Mac); it is fixed in versions 7.4.8 and 7.2.15 and there is no virtual patch available. Fortinet reports no active exploitation of any of these vulnerabilities.

It is recommended that you prioritise exposed FortiMail and FortiOS systems and identify devices in your fleet where TCP port 9191 is listening or the FortiVRS.exe process is running, so that these can be updated first.

More info

Google patches an actively exploited 0-day vulnerability in Chrome

Google has released a security update for Chrome that patches nine vulnerabilities, including CVE-2026-125350 (which has not yet been listed on the NVD but is classified as high severity by Google), a use-after-free vulnerability in Chrome’s XML component for which, according to Google’s advisory, an exploit is currently in circulation.

The flaw was reported by an anonymous researcher on 6 November, and Google has not provided details regarding attackers, targets or the delivery vector, nor has it confirmed whether it allows code execution. The update also fixes six other high-severity flaws, including a second ‘use-after-free’ in XML and a type confusion in V8, with no known exploitation. The patched versions are 157.0.8103.50/.51 for Windows and macOS, and 157.0.8103.50 for Linux.

Given Chrome’s widespread adoption and the phased roll-out of the update, it is recommended that you force the update and restart the browser on managed devices, and verify the installed version.

More info

Fake security tools on GitHub infect over 2,000 machines with AsyncRAT and a stealer

Doctor Web has documented a campaign that distributes malware through GitHub projects posing as security tools, with more than 2,000 machines infected. The activity began about two years ago with Visual Studio projects related to blockchain games, bots and cheats, targeting cryptocurrency wallet users. It later evolved into purported automated scanning, security testing and log analysis utilities, including one that impersonates a tool for processing Fscan scanner results.

The malicious code is hidden in pre-build commands and in PowerShell commands within configuration files. When the project is opened, a PowerShell script retrieves, from attacker servers and legitimate platforms (YouTube, Telegram, Google Docs, GitCode and Pastebin), an encrypted string pointing to an archive. The archive deploys Microsoft.exe, a modular Electron-based loader. The loader sends system information to a Telegram bot, escalates privileges, disables Windows Defender, UAC, System Restore and shadow copies, and removes McAfee, ESET and Malwarebytes.

The payloads include AsyncRAT, a clipper that swaps wallet addresses and takes screenshots, and a stealer that collects browser data, FileZilla, MobaXterm, Total Commander, VPN and messenger data, as well as Discord tokens.

More info

Claude Code’s “don’t ask again” option exposes 2,539 secrets in public GitHub repositories

Apiiro researchers Shmulik Cohen and Matan Giladi have identified 2,539 distinct secrets across 1,994 public GitHub repositories, exposed through Claude Code’s “Yes, and don’t ask again” option. When it is selected, the approved command is stored as a permission rule in .claude/settings.local.json, and any credential included in the command is saved in plaintext inside the project directory.

After examining 104,185 local settings files across 103,617 repositories, the researchers found 1,379 distinct secrets in 1,122 files, including 379 database passwords, 122 Supabase service-role keys, 103 Supabase account tokens, 43 SSH credentials and 23 AWS keys. In addition, 734 secrets appeared only in historical file versions. 63 organizations are affected. Gitleaks and TruffleHog missed 215 of the affected files, because passwords passed after PGPASSWORD=, sshpass -p or mysql -p have no recognizable signatures. Claude Code attempts to exclude the file through the developer’s global gitignore, but that configuration does not travel with the repository or protect files that are already tracked.

Users are advised to rotate credentials, review Git history, run git rm --cached on the file and add it to .gitignore.

More info

Operation Snowball: an autonomous AI agent compromises a European telecoms operator for two weeks

Malbear Labs has documented Operation Snowball after discovering, in a public directory, the complete workspace of an operator who, for more than two weeks, used an autonomous intrusion harness against a European telecoms provider and 13 organisations hosted in its data centres and cloud.

The agent ran on DeepSeek Flash, with its safeguards bypassed by a 78,000-character system prompt based on the “Little Code Sauce” jailbreak template, and selected CVEs, tools and methods autonomously; the instructions were in Chinese, although analysts emphasise that this does not prove nationality or state sponsorship.

The primary target was Cisco BroadWorks XSP, where it exploited CVE-2026-20320 (CVSSv3 7.5 according to Cisco), an unauthenticated XXE vulnerability in the OCI XML parser on port 2208, to extract the system’s private identity key and other credentials. Notably, the attackers adapted to the defences: faced with blocks from Trend Micro and OfficeScan, they changed their escalation technique, wrote their own HTTP-TCP bridge in Perl, and switched to persistence using only PowerShell, WMI and scheduled tasks, with post-exploitation via Sliver after bypassing Cloudflare.

It is recommended not to expose port 2208 and to monitor endpoint and network telemetry for patterns of automated scanning and lateral movement.

More info

◾ This newsletter is one of the deliverables of our Operational and Strategic Intelligence service. If you are interested in knowing the rest of the Operational and Strategic Intelligence contents included in the service, please contact us →