Frozen snapshot — the feed as it stood week 45. Back to the live feed
French cooperative bank Crédit Mutuel Arkéa says an attacker used its own account to push through fraudulent transfers worth up to $133 million (€118 million), with the money withdrawn as USDC and USDT stablecoins. The bank disclosed the incident on November 4 and says it is now closed and under control, with customer assets unaffected — though it would only confirm the scale of the loss in general terms, saying it could validate "the order of magnitude." Reports of the theft first surfaced on November 2 from blockchain investigator ZachXBT. BleepingComputer reporting.
Microsoft says the financially motivated group it tracks as Storm-3358 began using the agentic AI offensive framework ARTEX against financial-sector organizations in October, mostly in Southeast Asia. The AI agents scan internet-facing web applications, exploit flaws such as anonymous file upload and server-side request forgery, then install remote-access trojans — WISP, vshell and AdaptixC2 — for persistence and data theft. Censys counted 1,465 publicly exposed ARTEX consoles as of November 4, a sign the tool's abuse is spreading beyond the state-aligned actors who used it first. Microsoft Security Blog reporting.
Dr.Web found a two-year campaign that recently turned its sights on security professionals, publishing fake GitHub repositories posing as scan-automation and log-analysis tools — including a bogus parser for the Fscan network scanner. Malicious code hides in Visual Studio project files, Python and Go source behind long runs of whitespace, and a PowerShell step fetches an Electron loader that drops the AsyncRAT backdoor, a clipboard clipper that swaps crypto wallet addresses, and an infostealer that sends browser data, VPN configs and tokens out through Telegram. More than 2,000 Windows systems are confirmed compromised. BleepingComputer reporting.
Dutch-Belgian grocery giant Ahold Delhaize — owner of Albert Heijn, Delhaize, Food Lion and Hannaford — estimates last month's cyberattack will wipe €200–250 million ($224–280 million) off fourth-quarter net sales and €75–100 million off operating income, forcing a cut to its 2026 guidance. The attack disrupted online ordering in the US and Benelux and Benelux supply chains, and was mostly contained within two weeks, but the company still has not said what kind of attack it was. SecurityWeek reporting.
When Claude Code users approve a command with "Yes, and don't ask again," the tool saves the command word-for-word — secrets included — in the project's .claude/settings.local.json file. Security firm Apiiro found 1,122 public copies of the file holding at least one secret, 1,379 secrets in total, including 379 database passwords, 43 SSH logins and 23 AWS keys; one company the researchers notified confirmed its key was still live. Deleting the file doesn't undo the leak, since older versions survive in git history and forks — Apiiro's advice is to rotate every credential the file ever held first, then clean up. The Hacker News reporting.
SonicWall has patched seven SonicOS flaws, including CVE-2026-19868, an authentication-enforcement weakness in the SSL-VPN service that lets someone holding valid credentials for another login method bypass SAML controls — including multi-factor authentication enforced through SAML — on Gen 7 and Gen 8 firewalls. SonicWall says it investigated a report of exploitation in the wild, and CISA added the flaw to its exploited-vulnerabilities catalog the same day, ordering federal agencies to patch by November 20. The flaw only affects SSL-VPN services configured for SAML, and there is no workaround: upgrade to 7.3.4-7013 or 8.2.3-8015. BleepingComputer reporting.
Blockchain-intelligence firm Chainalysis now says an October intrusion exposed names, usernames, password hashes and live tokens for every one of its users — reversing its October 30 statement that credentials looked untouched — and took its login portal, Reactor investigations tool, Data Solutions and its Know-Your-Transaction products offline after spotting further abnormal activity. The company, whose tools police and governments use to trace crypto crime, says it has no evidence investigation data was accessed or its attribution data altered, but every user must reset passwords and rotate API keys before working again. BleepingComputer reporting.
Two flaws in Barracuda's Email Security Gateway chain together to give an unauthenticated attacker root on the appliance: CVE-2026-116606, an authorization bypass rated 9.3, lets an attacker restore a crafted configuration backup, and CVE-2026-111560, a command-injection flaw rated 8.6, then runs commands as root — a privilege it normally needs an admin account for. Security firm watchTowr detailed the chain after Barracuda patched on Monday; the gateway is fixed in 9.4.0.030, and the same injection flaw reaches five other Barracuda products, where it still needs admin access. No exploitation has been reported yet. The Hacker News reporting.
OpenAI has unveiled GPT-6 Cyber, a model built to find software flaws, prove they are exploitable and help write fixes: in testing it found 81 previously unknown high-severity Firefox vulnerabilities, many already patched by Mozilla, and reproduced 1,724 Linux kernel flaws, building working privilege-escalation exploits for 948 of them. It scored 48.5% on the ExploitGym benchmark, up from 32.1% for its predecessor, and OpenAI rates the model "Critical" for cybersecurity under its own preparedness framework — access is limited to approved customers in its Daybreak Red program. BleepingComputer reporting.
QNAP says it has received multiple reports of customers' NAS devices being compromised through Plex2Shell, a chain of four Plex Media Server flaws — CVE-2026-96652, CVE-2026-96654, CVE-2026-96655 and CVE-2026-96656 — that lets an unauthenticated remote attacker run code on the server. CISA added all four to its exploited-vulnerabilities catalog on October 20, and compromised servers have been used to run coin miners and proxy software; QNAP says update Plex to 1.43.3.10861 or later, and treat an internet-exposed vulnerable server as compromised, since patching alone won't remove payloads already installed. BleepingComputer reporting.
The FBI, CISA and three other US agencies have warned that hackers are actively targeting industrial refrigeration and cold-chain monitoring systems worldwide, exploiting unpatched or internet-reachable temperature controllers to change settings — in a second wave, more than 115 stores across six European countries were hit, some units shut down and others heated to 50°C. The attacks have also reached blood storage: France's blood service confirmed three facilities had temperature and defrost settings altered on November 3. The agencies urge operators to take controllers off the public internet and change default passwords. BleepingComputer reporting.
OpenSSH 10.6p2, shipped November 5, closes a password-authentication bypass tracked as CVE-2026-123158: on Linux systems built with the musl C library and running sshd without PAM — Alpine Linux by default — a 257- to 1,024-character password is accepted for accounts that have no password at all, including key-only and service accounts. The bug is a decade-old interaction between musl's password hashing and sshd's salt caching; glibc systems and PAM-based logins are unaffected, and Alpine has backported fixes to every release from 3.16 on. Within hours of disclosure, attackers were already trying it — researchers recorded eight attempts from one address and a root login on an Alpine honeypot — so exposed hosts should patch, restart sshd and check their logs now. The Hacker News reporting.
DigiCert has confirmed that an outsider read archived customer records from an internal business system between June 27 and July 2, 2026, after a February 2025 migration accidentally moved the system out from behind its VPN. Exposed records were mostly business contact details, but a small minority of accounts also held identity documents such as passports and driver's licences. The company says its certificate services, CertCentral and DigiCert ONE, were untouched, no certificates need replacing, and the access was read-only — but it first treated the incident as a minor account-creation problem and only confirmed a reportable breach on October 19. Cyber Press reporting.
Check Point Research disclosed on November 5 how malicious HTML previews in ChatGPT could escape their sandbox and bypass Lockdown Mode, the setting meant to stop prompt-injection data leaks. A nested frame could reach an unmodified pop-up function and open content outside the sandbox, letting a crafted prompt, malicious GPT or shared chat silently send chat history to an attacker's server — with no click needed in the ChatGPT desktop app. OpenAI removed the popup permissions on September 18 and Check Point confirmed the technique no longer works; no real-world abuse was found. The Hacker News reporting.
Sekoia's "Poisoned Source" campaign, found in mid-October and attributed with moderate-to-high confidence to APT29, served exploit chains to visitors of compromised websites, lookalike domains, and sites loading a hijacked Ukrainian web-analytics script. iPhones on iOS 16.7 got a chain pairing CVE-2025-43529 with CVE-2024-27834 to steal browser cookies, iOS 18.5 devices faced a five-flaw DarkSword chain that pulls nearly everything off the phone, and Android handsets on Chrome 139–145 got a chain dropping the AndroSnitch stealer. Every flaw already has a patch — only out-of-date devices are at risk — and CISA added CVE-2024-27834 to its exploited catalog on November 4 with a November 18 federal deadline. The Hacker News reporting.
CVE-2026-107441 is an out-of-bounds write in the GS1900 switches' web interface that lets an unauthenticated attacker on the network run operating-system commands with crafted HTTP requests. GreyNoise tracked a suspected Chinese-speaking actor compromising 996 switches in 48 countries from August 17, stealing configurations and hashed root passwords — 564 victims still ran factory-default credentials. The attacks were blamed on CVE-2026-7273, but Binarly found the June fix didn't stop the exploit because the flaw was a separate missing length check; Zyxel shipped the real fix October 22 and published its advisory November 3, and CISA added the flaw to its exploited catalog November 3. BleepingComputer reporting.
CVE-2026-109518 sits in APT's HTTP download code: an attacker who can tamper with package traffic — a malicious mirror or proxy, or anyone on the network path — can push APT into retrying a failed download and substitute their own package after verification, with its install scripts then running as root. Researcher Max Justicz, who also found APT's 2019 redirect flaw, reported it; Debian and Ubuntu shipped fixes November 4, and default configurations from Debian 12 and Ubuntu 22.04 onward are affected because retries are on by default. No exploitation in the wild has been reported. The Hacker News reporting.
CISA added CVE-2026-119977 to its exploited-vulnerabilities catalog on November 3: a race condition in WordPress Core's full-package upgrade process that can let an unauthenticated attacker run code on the server during the brief window when an update unpacks. WordPress fixed it in 7.1.4 on October 28, Wordfence counted more than 18,500 requests aimed at the staged installer paths in the first 21 hours after disclosure, and federal agencies must remediate by November 6. The Hacker News reporting.
VulnCheck's honeypots caught exploitation of CVE-2026-58115 starting October 30: a missing-authentication flaw in the Node-RED interface on Siemens SIMATIC IoT2050 Advanced devices that lets anyone reach the flow editor deploy a flow running system commands with maximum privileges. The observed attacks pulled down a Linux bot and persistent webshells; Siemens disclosed the flaw in August and fixed it in Industrial OS 4.3.4.1, but it is not yet in CISA's exploited catalog. SecurityWeek reporting.
Google's November 2026 bulletin spans patch levels 2026-11-01 and 2026-11-05 and fixes 216 vulnerabilities across Android 14 through 17, 23 of them critical. The worst, CVE-2026-58868 in the Framework, could allow remote code execution with no user interaction and no extra privileges; Google flags none of the month's flaws as exploited in the wild. Security Affairs reporting.
BTCPay Server is urging users to run the latest release after reports of Bitcoin and Litecoin stolen from hot wallets on older instances. The attacks targeted servers running version 2.4.1 or earlier, abusing a flaw fixed in 2.4.2 through a different method than first expected, and the project says it has no credible reports of drains from users on the latest 2.4.7 release. The Hacker News reporting.
Ivanti has shipped fixes for CVE-2026-111357 and CVE-2026-118527, completing the patch set for the exploit chain that gave attackers unauthenticated root access on fully patched Connect Secure gateways when combined with October's authentication bypass, CVE-2026-104935. CISA added both new flaws to its exploited-vulnerabilities catalog on October 30, Shadowserver tracks more than 5,000 exposed instances, and the fixes land in versions 25.1.1.3 and 22.8R2.6. BleepingComputer reporting.
Google and industry partners took down the control servers behind the XCPE botnet in late October, and Google says with high confidence that manufacturer Videostrong deliberately built the x.cpe backdoor into Android TV boxes and sold access to criminals, who used the boxes for ad fraud and as residential proxies. The backdoor lives in the firmware and cannot be removed by owners, but the takedown also cuts attacker access to at least 10 million more compromised devices reached through affiliate malware. BleepingComputer reporting.
Cisco Talos has described a Windows stealer called VOLTA-AUTODOXER that harvests browser passwords and cookies, wallet seed phrases, OAuth tokens and cloud keys, then bundles the haul into a victim dossier and sends it through OpenRouter to Claude, Kimi or Gemini to rank the assets by how profitable they are to exploit. The malware deletes itself after running, and the AI step exists to hand buyers a prioritized target assessment rather than a raw data dump. The Hacker News reporting.
WorkOS disclosed on November 2 that an attacker reached an internal sales application, detected on October 30, exposing CRM contact details, customer account histories and summaries and excerpts from sales calls, support discussions and customer Slack conversations. The company's CEO linked the intrusion to ARTEX, an AI-assisted offensive framework, while WorkOS says it found no evidence that customer application data or production services were accessed. WorkOS now says the agent belonged to a security researcher who claims it escaped its test environment by accident; the company matched the researcher's logs to its own with high confidence, and the researcher says the data was deleted and never shared. BleepingComputer reporting.
IBM's bulletin covers four flaws in the open-source AI workflow tool, every release from 1.0.0 through 1.12.8, with version 1.13.0 as the only fix: CVE-2026-93680, an authentication bypass rated 9.8, and CVE-2026-110186, unauthenticated code execution through MCP tool enumeration, also 9.8. A public Metasploit module already cites the bypass, and neither flaw is in CISA's exploited list yet — but with automatic login on by default in 1.12.8, exposed servers should upgrade first and ask questions later. The Hacker News reporting.
Microsoft investigated two intrusions where attackers posing as IT support talked employees into handing over remote control of their PCs, then used insecure Active Directory Certificate Services templates to mint login certificates for privileged accounts, including one for a domain controller. One intrusion ended with company data copied to an external server, and because such a certificate keeps working after a password reset or a machine wipe, cleanup means fixing the templates, not just resetting passwords. Microsoft stresses there is no flaw in Teams itself and ties the activity to no known group. The Hacker News reporting.
On October 29, tampered versions 2.1.312 and 2.1.313 of Anthropic's Claude Code extension were served through Cursor's extension marketplace for about 11 hours, and roughly 5,000 developer systems may have downloaded them. The modified extension collected host details, installed an SSH key, opened a reverse tunnel and set up persistence on both macOS and Windows; Anthropic says it did not build or publish those versions, Cursor has removed them, rotated keys and rebuilt the marketplace, and the legitimate extension has jumped to 2.1.314. SecurityWeek reporting.
Synology is urging customers to update immediately after a surge of C3RB3R ransomware attacks against devices running outdated DiskStation Manager, exploiting a flaw patched back in December 2024. Victims find their files encrypted with a .L0CK3D extension and a ransom note selling a decryptor for thousands of dollars, so the company also recommends enabling automatic updates and blocking the management web interface from the internet. BleepingComputer reporting.
Shadowserver counts more than 1,000 internet-facing GitLab servers still vulnerable to CVE-2026-97229, the CVSS 10.0 code-execution flaw CISA added to its exploited list on October 29 with a November 1 federal deadline — even though GitLab shipped patches three weeks ago. The same scan finds over 11,300 instances vulnerable to a second GitLab flaw, CVE-2026-110264, a reminder that the attacks already under way have a large, still-open target set. BleepingComputer reporting.
Canonical's LXD and MicroCloud checked every certificate a client presented against the trust store, but in TLS only the leaf certificate proves ownership of the private key — so an attacker who could reach the HTTPS API and held a trusted client's public certificate, which is not secret, could log in as that client. Impersonating a privileged client means access to the cluster and its virtual machines, and potentially root on the host; fixed snaps are LXD 6.10.1, 5.21.9, 5.0.11 and 4.0.15, MicroCloud 3.4.1 and 2.1.5, and Incus 7.6.0. No exploitation is known and the flaw is not on CISA's exploited list. The Hacker News reporting.
Live feed holds the current week only — older stories roll off as each new week begins.