Security
SonicWall SSL-VPN zero-day exploited as CISA orders fixes
SonicWall patches an exploited SSL-VPN auth bypass, CISA flags a WordPress core flaw, and a fake GitHub tool campaign has infected 2,000 Windows systems.
This edition was produced with artificial intelligence. Text and voice are generated automatically.
SonicWall fixes exploited SSL-VPN authentication bypass in SonicOS
SonicWall released updates for seven SonicOS firewall vulnerabilities, including CVE-2026-19868, an SSL-VPN authentication bypass exploited in attacks. The flaw, rated 7.1, lets attackers with valid credentials for an alternate authentication method bypass SAML controls, potentially including SAML-enforced multi-factor authentication. It affects only SSL-VPN services configured for SAML. CISA added the CVE to its Known Exploited Vulnerabilities catalog, confirming in-the-wild exploitation, and ordered federal civilian agencies to secure systems by November 20, 2026. The advisory also fixes two management interface flaws rated 8.8, pre-auth buffer overflows in SSL-VPN and TACACS+ services, and weak password hashing on virtual appliances. Affected versions are SonicOS Gen7 7.3.3-7015 and earlier and Gen8 8.2.2-8015 and earlier; fixes ship in 7.3.4-7013 and 8.2.3-8015, with no workarounds available.
SonicWall warns of new SSL-VPN MFA bypass flaw exploited in attacks →
Topics: Zero-day vulnerabilitiesPatch Tuesday & security updates
CISA confirms active exploitation of WordPress core race condition
CISA added a WordPress Core vulnerability to its Known Exploited Vulnerabilities catalog after evidence of active attacks. The high-severity flaw, CVE-2026-119977, carries a CVSS score of 8.1 and is a time-of-check time-of-use race condition that could let unauthenticated attackers execute code on the server during the upgrade window. CISA attributed it to predictable, web-accessible staging paths used during full-package core upgrades. Patches were released late last month. Wordfence said exploitation needs no site account but requires hitting a brief window during a complete Core update; it recorded more than 18,500 requests targeting staged setup-config.php installer paths in the first 21 hours after disclosure, and blocked 12,632 exploitation attempts in the previous 24 hours. Federal civilian agencies have until November 6, 2026, to apply fixes.
CISA Confirms Hackers Exploiting WordPress Core Flaw During Software Upgrades →
Topics: Zero-day vulnerabilitiesPatch Tuesday & security updates
Cisco Snort RCE flaw triggers on ZIP archives in HTTP responses
Trend Micro published advisory ZDI-26-867 on November 3, 2026, documenting a remote code execution vulnerability in Cisco Snort, the widely deployed intrusion detection and prevention system. The flaw, CVE-2026-20043, sits in the sf_unfold_header function and triggers while inspecting HTTP responses containing ZIP archives. It fails to validate user-supplied data, producing an out-of-bounds write (CWE-787) that executes code in the context of the Snort service account, with no authentication required. The official CVE record assigns a CVSS 3.1 score of 9.0 critical, while the ZDI advisory lists 9.8, a discrepancy the sources do not explain. Cisco has released a corrective update. The sources document no in-the-wild exploits and do not specify affected versions or fix details, so organizations must consult Cisco for compatibility.
Snort, the Network Guardian, Falls to a ZIP Archive | DeafNews →
Topics: Patch Tuesday & security updates
Fake GitHub security tools infect 2,000 Windows machines with backdoors
A campaign distributing malicious GitHub repositories has compromised more than 2,000 Windows systems, according to researchers at Dr. Web. The campaign began about two years ago, initially targeting crypto wallet users through blockchain games, bots and cheat tools, and has recently shifted to cybersecurity professionals with repositories advertised for automating scans, security testing and log analysis. One identified repository posed as a lightweight script for parsing output from the Fscan network scanner. Malicious code was hidden in Visual Studio project files, Python scripts and Go source files behind long stretches of whitespace. The infection chain uses PowerShell to fetch a modular Electron-based loader named Microsoft.exe, which establishes persistence, elevates privileges, disables Windows security and recovery features, and injects the next stage into legitimate processes using six evasion techniques. Final payloads include the AsyncRAT backdoor, clipboard-hijacking clipper malware and an infostealer that exfiltrates data through a Telegram bot.
Fake security tools on GitHub push backdoors and infostealers →
Topics: Malware & infostealersSupply chain attacks
Alpine Linux OpenSSH flaw allows passwordless logins
Administrators running sshd on Alpine Linux servers should update openssh and restart sshd immediately, since upgrading the package alone leaves the old daemon running. A flaw in the OpenSSH server lets attackers log in without a password using a single connection that works every time. The attacker needs the name of an account whose password field is an asterisk, normal for key-only accounts, plus at least one other account with a $5$ or $6$ hash, then sends a password of 257 to 1,024 bytes. When built without PAM, sshd hashes each attempt with crypt() and compares it to the stored field, borrowing another account’s salt when that field is too short; musl’s crypt() returns an asterisk as an error indicator for keys over 256 bytes, matching the stored field of passwordless accounts. OpenSSH 10.6p2 fixes it, with no CVE assigned at the time of writing. Every Alpine release checked is affected, and fixed packages ship back to 3.16.
Alpine Linux SSH Flaw Lets Attackers Log In Without a Password →
Topics: Patch Tuesday & security updates
Claude Dispatch flaw lets browser extensions run commands on the OS
A browser extension with permission to control claude.ai could use Claude Dispatch to make Claude Desktop execute commands on a machine, including a different machine signed into the same Claude account, without a browser exploit. Dispatch’s interface and API were reachable from an ordinary authenticated claude.ai session; once the URL was extracted from the desktop app, the page could be opened in a browser tab and controlled through the logged-in session. In a proof of concept, a content script opened the Dispatch agent page in an unfocused popup, obtained the local Desktop session ID using ambient cookies, and sent prompts sequentially; a script clicking any Allow button handled approvals, and a package preinstall hook opened Calculator, demonstrating local code execution. The researchers call the technique ThisBash. Anthropic acknowledged the report on May 19, 2026, closed it as a duplicate, and said Dispatch would be retired targeting the end of October 2026.
Topics: ClaudeAI coding agents security
macOS 27.3 removes eight compromised eMudhra certificates
Apple is addressing a certificate security vulnerability in macOS 27.3 affecting developers and other Mac users who rely on command-line tools. The fix targets eight compromised intermediate certificates issued by Indian certificate authority eMudhra, removing them from a system-level certificate store used to establish secure connections. Apple already blocked the affected certificates for Safari, Mail and apps using its native networking frameworks on October 19, so most Mac users do not need to wait for the update. A researcher found that private cryptographic keys associated with the eight certificates were weak enough to raise security concerns; if attackers obtained or reproduced them, they could potentially impersonate legitimate websites and intercept sensitive information. There is currently no evidence of exploitation. The remaining gap involves the /etc/ssl/cert.pem file used by some Terminal-based tools, which the macOS 27.3 beta cleans up. The public release is expected in December.
Topics: Apple
North Korea grooms child prodigies into crypto thieves
North Korea selects 200 of its smartest middle school students each year to study computer engineering, with a fraction eventually becoming hackers, according to defectors and analysts. The programme is more than a decade old and was assembled by leader Kim Jong-un. Top hackers are rewarded with perks including spacious apartments in the capital and large hard currency awards for gruelling hours of work. The group has evolved from running illegal gambling websites in its early days to carrying out the world’s biggest cyber heist last month, the Binance crypto hack, which netted US$6.3 billion.
How North Korea turns child prodigies into crypto thieves →
Topics: State-sponsored hackingCrypto security
GhostMap Android flaw forces persistent boot loops without permissions
A malicious Android app requiring no permissions and no launch can force a device into a persistent boot loop by exploiting a vulnerability Bitdefender named GhostMap. The flaw makes Android map a file into an unallocated memory region; the affected process crashes when it reads from that region. Bitdefender discovered it while analyzing an Android banking trojan that kept crashing its emulators when reading an application resource. GhostMap stems from two Android libraries interpreting ZIP compression methods differently, allowing a malformed APK to bypass size checks and be mapped beyond the file’s actual contents; reading such an entry causes a SIGBUS in whichever process touches the file. Bitdefender’s demonstration app contains a resource that the core Android service system_server automatically reads after installation, triggering an endless loop. Testing used a Pixel 10 running Android 17 with the August 5 security patch, looping 110 times in normal boot mode and 72 times in safe mode. Recovery requires a factory reset or ADB deletion. Bitdefender reported the issue to Google on August 20 and received a Won’t Fix (Not Reproducible) reply deeming it out of scope.
GhostMap Android flaw lets malicious app trigger persistent boot loop →
Topics: Zero-day vulnerabilities
Firecracker guest-to-host escape achieves root code execution
A guest-to-host escape in Firecracker achieved code execution as root on the host by hijacking the saved userspace register frame of a live vCPU thread. The work demonstrates CVE-2026-99999 on a Firecracker setup, continuing from the host memory primitive introduced in PwnEscape Part 1. Each Firecracker vCPU runs as a thread in a host process that enters KVM via KVM_RUN and returns to userspace to handle events; while the thread is in that system call, the host kernel holds its saved userspace registers in a pt_regs structure whose instruction pointer and stack pointer determine where execution resumes. Modifying that frame lets an attacker choose the code Firecracker runs on return from KVM. The chain used a nonce in a guest page to find the EPT root and recover the guest page’s host physical address, then a reverse mapper to recover the Firecracker process’s page table root, an image mapper to defeat ASLR, and a frame finder to locate the pt_regs frame. The payload made Firecracker write an eight-byte string to its own stderr, with all three modified stack words restored afterward. The target was Firecracker v1.16.1 on x86-64, host kernel 7.0.0-1011-aws, launched directly as root under systemd-run with no jailer.
Firecracker VM Escape: PwnEscape Part 2. | PWN.AI Security Research →
CPython zero-day enables sandbox escape in AI serving stacks
RunSybil discovered a zero-day memory-safety vulnerability in the CPython interpreter that allows attackers to escape a code sandbox and execute code hidden in a specially crafted model, enabling breaches of AI hosting platforms. Exploitation could give attackers a foothold inside servers processing millions of user requests and strike at underlying cloud infrastructure. The flaw abuses CPython’s handling of generic types: CPython caches an alias’s blank parameters in a tuple, and crafted objects with typing_subst and typing_prepare_subst hooks let an attacker modify that cache after its bounds are checked. When a parameter arrives after caching, a lookup returns -1; the assertion guarding it is compiled out in release builds, so CPython reads argitems[-1] and passes the result to the attacker’s callback as a Python object pointer. Starting in CPython 3.14, a tuple’s cached hash sits immediately before its element pointers, so a deterministic tuple-of-integers hash can be chosen to equal a desired address. The researchers recovered the live str class, then the type and code classes, and applied type to Jinja’s safe_range wrapper to obtain constructors for building and calling a code object. They achieved code execution in a Jinja ImmutableSandboxEnvironment and, using a Qwen model with a backdoored chat_template.jinja, executed arbitrary code in a vLLM serving process during chat rendering. CPython said it does not support sandboxing untrusted Python code as a security boundary; Jinja’s maintainers said to report it to CPython; Hugging Face did not respond within the 90-day window; vLLM acknowledged the report but has not decided on a mitigation. DigitalOcean’s isolation contained the attack and AWS implemented a mitigation in Bedrock.
CPython 0-Day Vulnerabilities Threaten the AI Supply Chain | RunSybil Blog →
Topics: Zero-day vulnerabilitiesAI infrastructure & data centers
Masscan sweep finds 1.1 million hosts answering on TCP port 0
Port 0 is reserved by IANA and treated by the sockets API as a request for an ephemeral port, but nothing on the wire prevents a TCP packet from being addressed to it. RFC 9293 defines the port field as 16 bits with no rule against zero; RFC 6335 lists 0/tcp and 0/udp as Reserved, set aside rather than forbidden. On Linux, bind() with sin_port = 0 requests any port, drawn from the ephemeral range without privileges, and the kernel splits the range by parity, allocating odd ports for listeners and leaving even ports to outbound connections. Tests on Linux 6.8 confirmed 2,000 bind() calls returned odd ports and 300 connect() calls returned even source ports. An IRC network was placed on port 0 using iptables REDIRECT rules; most clients refuse port 0, though socat and Python asyncio connect. A masscan sweep of all IPv4 for port 0 drew SYN-ACKs from 1,173,626 addresses, concentrated in a few thousand /24 blocks, and 1,116,675 addresses completed the handshake without sending data. Of 56,951 that sent a banner, 53,741 were F5 BIG-IP load balancers writing reset reasons into RST packets. After filtering, 2,905 addresses remained, exposing 748 SSH, 215 MySQL/MariaDB, 211 telnet, 159 HTTP, 105 FTP, 80 SMTP, 44 DVR/camera, 39 VNC and 21 OpenVPN services. The most sensitive exposure was 21 Avaya IP Office systems streaming SMDR call records to any connecting client, roughly 7.5 KB per host covering 37 to 62 calls with phone numbers, staff names, extensions and internal addresses; 19 of the 21 are in the UK.