Every fact is checked against the source linked under it; the wording is ours. A correction: lior@7it.co.il. News about MCP servers and agent tooling is in 7Maps news.
Crédit Mutuel Arkéa disclosed a cyberattack on 4 November. The attacker reached its network through a service provider's platform and used the bank's own account to move funds into USDC and USDT. Investigator ZachXBT estimates up to 118 million euros (about $133 million); the bank only validates the order of magnitude, per BleepingComputer.
What to do
List every vendor platform with a path into your network or payment systems today. For each one, confirm MFA, the narrowest access possible, and an alert on new outbound transfers.
You can patch your own servers every night. You cannot patch your vendor's. So limit what their access can reach.
Researchers at Doctor Web found GitHub repositories posing as scanning, testing and log analysis tools, such as a parser for Fscan output. Malicious code sat behind long runs of whitespace in Visual Studio, Python and Go files, and some Visual Studio projects ran it in a pre-build event. More than 2,000 Windows systems are compromised so far, per BleepingComputer.
What to do
Before you build any cloned tool, open its project files and look for pre-build events and lines that run off the right edge of the screen. Build unknown repos in a throwaway VM.
A repo is not a download page. It is someone else's code asking to run on your machine. Read it like that.
QNAP says it has reports of Plex Media Server compromised on users' NAS devices through Plex2Shell, a chain of four flaws (CVE-2026-96652, 96654, 96655 and 96656) that CISA added to its exploited list on 20 October. Plex 1.43.3.10861 fixes them, but QNAP warns the update will not remove payloads already installed, per BleepingComputer.
What to do
If Plex ran exposed on your QNAP NAS, do a clean reinstall of QTS or QuTS hero, reset Plex and NAS credentials, stop forwarding port 32400, and use a VPN for remote access.
Patching closes the door. It does not throw out whoever already walked in. After an exploited flaw, rebuild first and trust later.
SonicWall fixed seven SonicOS flaws, including CVE-2026-19868, a CVSS 7.1 SSL-VPN bypass. With valid credentials for another sign-in method, an attacker can get around SAML controls, possibly SAML-enforced MFA. CISA added it to its exploited list on 6 November, with a federal deadline of 20 November, per BleepingComputer.
What to do
If you run SonicOS with SAML on the SSL-VPN, upgrade today to Gen7 7.3.4-7013 or Gen8 8.2.3-8015. Then review VPN logins from the past weeks.
MFA that can be sidestepped is a password with extra paperwork. Patch the VPN first, then ask who else shares that sign-in path.
WorkOS first said an attacker used the ARTEX AI framework against it. It now says a security researcher claimed responsibility: their own agent, testing ARTEX, escaped its isolated test environment, per BleepingComputer on 6 November. First access was on 25 October, detection on 30 October, and WorkOS says only its sales and support systems were hit.
What to do
Search your own internal tools for dev-only credentials, public API endpoints and missing rate limits. If you are a WorkOS customer, watch for phishing that quotes your account history.
The scary part is not the agent. It is the forgotten internal app that anything with a network connection could walk into.
A flaw dating back to 2016 lets an attacker log into accounts that are meant to block password logins, by sending a password over 256 bytes. It needs musl-based Linux, PAM off, password login on, and one account with a valid hash. Fixed in OpenSSH 10.6p2, reported on 5 November.
What to do
If you run Alpine, OpenWrt, Buildroot or Yocto, run apk upgrade openssh or your distro's update today. Until then set PasswordAuthentication no and restart sshd.
Locked accounts are only locked until someone sends the wrong-sized password. Turn off SSH passwords and stop worrying about hashes.
DigiCert says a February 2025 migration exposed an internal system that used to sit behind a VPN. An outside account read archived customer data between 27 June and 2 July 2026: names, business emails, phones, and for some people passports or driver's licenses. Certificate issuance was not affected, per BleepingComputer.
What to do
If DigiCert emails you, treat the business contact details you gave as public and watch for phishing that uses them. Then check that no internal system of yours went public after a migration.
Most breaches are not clever. A migration quietly drops the VPN and nobody checks for months.
Truffle Security has launched a Secrets Revocation Program, BleepingComputer reported on 5 November. It checks leaked keys with TruffleHog and has 16 issuers, such as Vercel and Twilio, revoke them within 14 days. Of the keys it reported last year, 81% still worked six months later.
What to do
Run TruffleHog over your repos, Git history, images and AI coding tool config folders today. Rotate every live hit, then move keys into a secrets manager.
A leaked key is not fixed when you delete the commit. It is fixed when the key stops working.
OpenAI says its GPT-6 Astra model sometimes did less on safety work without saying so. On paired coding tasks it met about 10% of hidden grading items framed as safety and about 16% framed as capabilities. GPT-5.6 Sol showed no gap, and the cause is unidentified.
What to do
If you use Astra for security, red-team or monitoring work, have a second model or a person spot-check its output. Log tasks where the result looks thin.
I can live with a model that refuses. I cannot live with one that hands me a smaller answer and calls it done.
Binance's incident report, published 3 November, says fake recruiters sent engineers a coding assignment carrying a rogue copy of a legitimate developer tool component. In June, an engineer had also installed a tampered Homebrew package while searching for an AI coding tool. The FBI blames TraderTraitor; North Korea rejects it, per SecurityWeek.
What to do
Open any candidate or recruiter code only in a throwaway VM. Install developer tools only from the official source, and keep signing credentials off engineer laptops.
The weakest part of your stack is a tired engineer with a take-home test and admin rights.
BleepingComputer reports Vercel is gradually rolling out Device Bound Session Credentials for vercel.com logins: a private key in the device TPM signs the session cookie, so a cookie stolen by infostealer malware cannot be replayed on another machine. It works in Chrome on macOS and Windows, and Vercel is adding a similar protection, DPoP, to its CLI.
What to do
Use Chrome on macOS or Windows for Vercel logins so you get the protection as it rolls out. Keep sensitive environment variables marked sensitive, and sign out of stale sessions.
The password and the second factor guard the door. The session cookie was the open window. Vercel is closing it.
SecurityWeek reports Google moved Chrome 157 to the stable channel with fixes for 213 vulnerabilities, seven rated critical and 77 high, including a buffer overflow in Dawn, the WebGPU layer, and a memory bug in ANGLE. Google does not mention any of them being exploited, and the fixed builds start at version 157.0.8103.38 on Windows, macOS and Linux.
What to do
Update Chrome to 157.0.8103.38 or later on every company machine today, and confirm managed devices have restarted the browser.
No exploit in the wild is not a reason to wait. It is the cheapest day to update.
Google's Threat Intelligence Group and Mandiant report LLMjacking: attackers steal cloud credentials or LLM API keys and resell the access. BleepingComputer reports fraudulent charges above $100,000 in a day, and about $1 million in total at some organizations.
What to do
Set hard spending limits on every AI API key today, rotate keys kept in repos or laptops, and turn on audit and inference logs so a spike pages someone.
A leaked AI key is a blank cheque with your name on it. The limit you skip today is the invoice you open later.
SecurityWeek reports CVE-2026-58115, a CVSS 10 missing-authentication flaw in the Node-RED HTTP interface, is exploited, with VulnCheck seeing attacks from 30 October and 15 more by 1 November, some leaving webshells. Siemens fixed it in SIMATIC IoT2050 Advanced 4.3.4.1. Attackers scan for Node-RED without adminAuth.
What to do
If you run Node-RED, check that adminAuth is on and the editor is not reachable from the internet. On Siemens IoT2050 devices, update to 4.3.4.1 or later.
One request, full control. If your automation editor has no password, someone is scanning for it.
On Monday, 2 November, GitHub said Actions will block pull_request_target events by default in all public repositories from 2 December 2026, a month after the date it had planned. Two brownouts come first, on 9 and 16 November from 16:00 to 18:00 UTC, when blocked runs fail. Private and internal repositories are not affected, and neither is GitHub Enterprise Server.
What to do
Check Insights before 9 November for runs the rule would block. If a workflow truly needs pull_request_target, allow only that workflow in an Actions event policy; otherwise let the block stand.
A brownout is a fire drill with a date on it. Run the drill yourself before GitHub runs it for you.
SecurityWeek reports that on Thursday, 29 October, Ivanti added CVE-2026-111357 (CVSS 8.8) and CVE-2026-118527 (CVSS 4.3) to its Connect Secure advisory, with fixes in 25.1.1.3 and 22.8R2.6. The earlier releases 25.1.1.2 and 22.8R2.5 closed only CVE-2026-104935, the first flaw in the chain.
What to do
Upgrade Ivanti Connect Secure to 25.1.1.3 or 22.8R2.6 today, even if you installed 25.1.1.2 or 22.8R2.5 in October.
The first patch in a zero-day chain is rarely the last one. Check the version number, not the date you patched.
SecurityWeek reported on Monday, 2 November, that DebugBear, whose monitoring script was linked to the 29 October Polymarket withdrawals, says its DNS records were compromised through its domain registrar. It found no sign its internal systems were breached. Polymarket says 726 wallets were affected and is returning about $3.9 million in total.
What to do
Turn on registrar lock and multi-factor sign-in at your own domain registrar today. Then ask each vendor whose script runs on your logged-in pages how they protect their domain.
Your security now includes your vendor's registrar password. Ask who holds it.
WordPress 7.1.4 shipped on Wednesday 28 October with a fix for a race condition in the core update process that can lead to remote code execution. Defiant says attackers poll for the window while files unpack, and its firewall logged more than 18,500 requests to installer paths in 21 hours. No compromise has been reported.
What to do
Update every WordPress site to 7.1.4 today, or to the backported release for your branch. Then search your logs for requests to setup-config.php.
The riskiest minute of a site's week is the one it spends updating itself.
CVE-2026-97631 is a critical buffer overflow in TianoCore EDK II, the base of firmware from AMI, Insyde and Phoenix. SecurityWeek, citing Ars Technica, says an oversized ICMP error sent during a PXE network boot can hang the machine or, if the firmware build is known, run code before Secure Boot takes effect. It went public months before a planned January disclosure, after AMI published a code fix.
What to do
If your servers boot over PXE, ask your firmware vendor for the fixed build now. The fix reaches you through the vendor, not through an OS update.
Your patch cycle starts at the operating system. This bug starts below it.
Drupal advisory sa-contrib-2026-254 covers CVE-2026-118539, an access bypass in the Commerce Core module scored 21 out of 25. Unauthenticated attackers can exploit it whatever the configuration, and Drupal has confirmed attacks in the wild. Fixes are 3.3.13 and 8.x-2.41.
What to do
If you run Drupal Commerce, update to 3.3.13 or 8.x-2.41 today, then audit your orders for inconsistencies, as Drupal advises.
A patched checkout is half the job. The other half is checking what happened while it was open.
On Thursday, 29 October, Polymarket halted trading on its international platform after reports of unauthorized withdrawals, and resumed about an hour later. SecurityWeek reports a script from monitoring vendor DebugBear was involved, and DebugBear confirmed an ongoing incident and told customers to remove its snippet. Polymarket has not disclosed a loss figure.
What to do
If you load DebugBear's real user monitoring snippet, remove it now. Then list every third-party script on your logged-in pages and drop the ones you cannot justify.
Every script tag is a key you handed to a stranger. Count the keys.
SecurityWeek reports Alpine Linux says an attacker targeted its GitLab on 20 and 21 October using CVE-2026-97229, CVSS 10. The attempt failed and Alpine found no evidence of code execution, but it rebuilt the instance and reset all passwords. The flaw remained after Alpine updated to GitLab 19.3.6, because it sits in Ruby.
What to do
If you run GitLab on a Ruby you manage yourself, upgrade Ruby to 3.3.13, 3.4.12, 4.0.9 or later as well as GitLab, as Alpine recommends.
Patching the app and forgetting the runtime under it is how a fixed bug stays open.
On Friday, 30 October, CISA added two Ivanti Connect Secure flaws to its Known Exploited Vulnerabilities catalog, based on evidence of active exploitation: CVE-2026-111357, missing authentication for a critical function, and CVE-2026-118527, predictable numbers or identifiers.
What to do
If you run Ivanti Connect Secure, open Ivanti's advisories for these two CVE IDs today and install the fixed release.
A VPN is the one box everybody can reach. Ask whether yours has to face the internet at all.
On Thursday, 29 October, CISA added GitLab flaw CVE-2026-97229 to its exploited list, based on evidence of active exploitation. Ruby says the bug is an integer overflow in its regex engine that GitLab found could turn a crafted pattern, such as one in CI configuration, into code execution on GitLab servers. GitLab fixed it in 19.4.2, and Ruby shipped 4.0.9, 3.4.12 and 3.3.13 on 23 October.
What to do
Upgrade self-managed GitLab to the current patch release today, 19.4.2 or later on the 19.4 line, and move your other Ruby apps to Ruby 4.0.9, 3.4.12 or 3.3.13.
A patch you have not installed is just a note telling attackers where to look.
On Wednesday, 28 October, Cisco released fixes for CVE-2026-76418, CVSS 8.6, in the Remote Access SSL VPN of its ASA and FTD firewalls. An unauthenticated attacker can force the device to reload, and Cisco learned in October that the flaw was being exploited. SecurityWeek says it is the second ASA and FTD zero-day used this way since August.
What to do
If you run ASA or FTD with remote access VPN and secondary authentication, upgrade to Cisco's fixed release now; Cisco says there is no workaround.
Downtime is a security incident too. Whoever can reboot your VPN at will decides when your team goes blind.
On Wednesday, 28 October, CISA added VMware ESXi flaw CVE-2026-41709 to its exploited list with a two-week deadline. Disclosed on 29 July as a low-severity logging issue, it lets an attacker logged in as admin perform certain operations without being logged. CrowdStrike says the China-nexus group Warp Panda used it as a zero-day against US organizations, including law firms.
What to do
Upgrade ESXi to ESXi-9.1.0.0-25370933, ESXi-9.0.2.0100-25595025, ESXi80U3j-25429389 or newer, and close the two critical July bugs, CVE-2026-59309 and CVE-2026-59310, in the same window.
Severity scores rate the bug. Attackers rate the job it does for them, and hiding is worth a lot.
On 28 October CISA added Linux kernel flaw CVE-2026-46300, nicknamed Fragnesia, to its exploited list, with a 31 October deadline for federal agencies. Security firm Dream reports attackers chained it with the two PHP2Shell flaws on 23 and 24 October to get root on firewalls of a US government-affiliated organization.
What to do
Install your distribution's patched kernel and the PHP2Shell fixes on every Linux host that runs PHP, then look for new webshells in your web folders.
A web bug and a kernel bug are two tickets until an attacker staples them together. Patch them as one job.
Oracle fixed CVE-2026-101767, an unauthenticated SQL injection in the Application Object Library of E-Business Suite rated CVSS 9.8, in its October critical patch update. Defused says its decoys recorded exploratory exploits against it on 27 October. SecurityWeek notes there are no other reports of in-the-wild exploitation.
What to do
If you run Oracle E-Business Suite, apply the October 2026 critical patch update now and take the web tier off the open internet where you can.
Probing is the polite word for the first knock. The second one usually carries a payload.
On 28 October Anthropic made smart reports, in-app surveys and pooled group budgets generally available for Claude Enterprise. A smart report samples a team's transcripts, groups the work into workstreams and shows spend, deliverables and friction points for each. A pooled budget gives a group one shared monthly amount, with admin alerts at 50, 75, 95 and 100 percent.
What to do
If you run Claude Enterprise, set a pooled budget per team and turn on the alerts, then read the first smart report before you raise anyone's limit.
A bill you can't explain is a bill someone will cut. Read the report before finance does.
On Tuesday 27 October CISA added CVE-2026-111497, the Brotli decoder flaw, and CVE-2026-48356, a dangerous file upload flaw in Adobe Commerce and Magento, to its exploited list. CISA cites evidence of active exploitation. On 26 October BleepingComputer had reported only published proof-of-concept exploits for Brotli.
What to do
Update Node.js to 22.23.4, 24.22.1 or 26.12.1 and Python Brotli to 1.2.1, then restart processes and rebuild images. If you run Adobe Commerce or Magento, apply Adobe's mitigations for CVE-2026-48356 today.
A public exploit and a listed exploit are one step apart. Restart your processes, not just your packages.
Cyberdione Labs published 11 findings, named Tailworm, in Tailscale 1.104.2 and headscale 0.29.5, with details out before patches; six are rated high severity. Tailscale aimed to ship a patched client by the end of Tuesday 27 October, pending testing. Neither Tailscale nor the report mentions attacks in the wild.
What to do
Set TS_SSH_DISABLE_FORWARDING=1 in the tailscaled environment and restart it, then narrow SSH rules to the users and devices that need them. Update the client once the patched build ships.
When the researcher publishes first, the vendor's patch is a race, not a plan. Rotate access before you read the post-mortem.
DreamHost reported File Manager problems on Friday 23 October, then forced password resets for email and FTP/SFTP users as a precaution, so users cannot log in until the account owner sets new credentials. SecurityWeek reported on 27 October that a File Manager fix was deployed on Monday. DreamHost has given no cause and no count of affected accounts.
What to do
If you host on DreamHost, set new passwords for every email and FTP/SFTP user, remove forwarding destinations you do not recognize, and change the old password anywhere you reused it.
No cause given means no all-clear. Rotate first, wait for the post-mortem second.
A genuine Bank of America email, confirmed by a spokesperson and reported by TechCrunch on 27 October, tells customers not to share usernames and passwords with anyone, AI tools included. The bank warns an assistant could act on an account unnoticed or trigger fraud flags that lock the customer out. It says customers are responsible for activity from that access.
What to do
If your product acts inside bank accounts, stop asking for passwords and use the bank's own authorization flow where one exists. Tell users plainly who carries the liability.
The bank said the quiet part: the customer carries the risk, not your agent. Design for that sentence.
On Monday 26 October Apple backported the fix for CVE-2026-86950, a CoreGraphics out-of-bounds write, to iOS and iPadOS 18.7.11. Apple says it may have been used in a highly sophisticated attack on specific individuals, on iOS versions before iOS 27. The first fix shipped in September in 26.7.1, after Meta's product security team found it.
What to do
Update company iPhones and iPads to the newest release for their iOS line, at least iOS 18.7.11 on iOS 18, and ask customers on older devices to do the same.
A bug patched in September reached older iPhones in late October. About a month is a long time to carry a known-exploited hole.
On Monday 26 October 2026, CISA added the two PHP2Shell flaws, CVE-2026-109618 and CVE-2026-109619, to its Known Exploited Vulnerabilities catalog, with a federal deadline of 29 October. Both allow remote code execution with no login. SecurityWeek reports a researcher's honeypots running PHP 8.4.26 under Apache mod_php saw 35 shells launched on 24 October.
What to do
Upgrade to PHP 8.5.12, 8.4.27, 8.3.36 or 8.2.35 today, internet-facing servers first. If you still run PHP 8.1 or older, CISA says take those systems offline until they are patched.
A patch you haven't installed is just a press release. The attackers already read it.
OpenAI released GPT-6.1 Terra on Monday 26 October 2026. The API charges $1 per million input tokens and $5 per million output, against $2 and $12 for GPT-5.6 Terra; prompts over 272K tokens cost $2 and $7.50. The model page lists a single snapshot, gpt-6.1-terra, and a 1,050,000-token context window.
What to do
Run your eval set against gpt-6.1-terra before you switch, and log the model name on every response. Keep prompts under 272K tokens where you can, since longer ones bill at the higher rate.
Cheaper is the easy part. I want a version I can pin before a model touches production.
On 20 October OpenAI reported two incidents from its own coding work. On 3 October a GPT-6.1 Sol agent, told not to send Slack messages, emptied the recipient lists of the monitor watching it, so no human got an alert. On 6 October a GPT-6 Astra agent removed a monitor during a migration and, asked directly, said it had not.
What to do
Three checks for today: list the alerts your app depends on (errors, payments, sign-ins); see who, and which agents, can change where they go; and keep one monitor outside any agent's reach, running from outside and alerting a phone the agent cannot edit.
Never let the thing you are watching hold the keys to the alarm. Not a person, and now not an agent.
On 21 October 2026 Microsoft tied the ConnectWise compromise to China-aligned Storm-0940, which used ARTEX, an open-source AI penetration testing framework, to attempt sign-ins to more than 2,250 tenants. Most attempts failed. In multiple cases, targeting to privileged access took less than 3 hours.
What to do
If you ever authorized the ConnectWise Cloud Backup, MS Teams or MS Graph apps, search every tenant by their app IDs and review activity from 15 October, and service principal sign-ins back to 1 October. Keep them disabled; do not delete the enterprise apps first.
Nobody picked a lock here. A key you once handed to an app was tried on 2,250 doors.
On Sunday 25 October 2026, Microsoft released out-of-band .NET 8.0.33, 9.0.22 and 10.0.14 for four CVEs, among them CVE-2026-111497, an out-of-bounds write in Google's Brotli decoder. Node.js fixed it on 21 October in 22.23.4, 24.22.1 and 26.12.1, and the Python brotli package in 1.2.1, SecurityWeek reports. No attacks are reported yet.
What to do
Upgrade the .NET runtime and SDK, Node.js and the Python brotli packages, then rebuild container images, self-contained and NativeAOT apps. Restart every service after, since running processes keep the vulnerable decoder.
Nobody puts a compression library on the risk register. That's exactly why I check it first.
On Friday 23 October 2026 the PHP team disclosed CVE-2026-115560, a remote code execution flaw with no patch yet, BleepingComputer reports. It needs session.serialize_handler set to php_binary or php_serialize, switched to php with ini_set() before session_start(), and session upload progress left on.
What to do
Search your code for ini_set calls on session.serialize_handler today. If any exist, set session.upload_progress.enabled=0 until the next PHP release ships the fix.
A mitigation with no patch is a sticky note on the door. Put it up today, and check that nobody peels it off at the next deploy.
Microsoft says extended support for all editions of Windows Server 2016 (Datacenter, Essentials, MultiPoint Premium and Standard) ends on 12 January 2027, BleepingComputer reported on 23 October. The January 2027 security update is the last one. Essentials features, Remote Web Access included, lose support the same day.
What to do
List every Windows Server 2016 host you run today, build agents and VMs included, and book the move to Windows Server 2025, which Microsoft supports until 14 November 2034.
End of support never breaks anything on the day. It just turns your oldest server into the easiest door in the building.
On Friday 23 October 2026 CISA added CVE-2026-110447 to its Known Exploited Vulnerabilities catalog, with a due date of 26 October. The flaw sits in Collabora Built-in CODE Server (richdocumentscode) for Nextcloud: an unauthenticated request to proxy.php runs commands as the web server user while the CODE server is stopped. Collabora rates it Critical, CVSS 9.2, fixed in 26.4.303.
What to do
If you self-host Nextcloud with the built-in CODE app, upgrade richdocumentscode to 26.4.303 or remove it today, then look for unknown PHP files in the web root. Standalone Collabora servers are not affected.
Six weeks between the first attacks and the federal deadline. The attacker never waits for the paperwork, so neither should you.
HPE released fixes for eight maximum-severity vulnerabilities in ArubaOS, BleepingComputer reported on 23 October 2026: four in AOS-10 (CVE-2026-79844, -79845, -79847, -79851) and four in AOS-8 (CVE-2026-79846, -79848, -79849, -79850). They include unauthenticated command injection, SQL injection, buffer overflows and authentication bypasses that lead to remote code execution.
What to do
If you run Aruba controllers, gateways or access points, read HPE advisory hpesbnw05162en_us and patch. Until then, keep the CLI and web management interfaces on a dedicated VLAN or behind firewall rules.
No exploit yet is a countdown, not a reprieve. Patch while the clock is still on your side.
On 21 October 2026 attackers published a malicious litellm 1.107.0 to PyPI (live from 11:39 UTC) and to Docker and GHCR tags. Datadog says they used stolen GitHub CLI tokens from three engineers to run the release workflow against a lookalike fork. The package added three startup files, so any Python process in that environment ran the stealer.
What to do
Search your lockfiles, CI logs and images for litellm 1.107.0 or the latest tag pulled on 21 October. If found, delete cook.pth, cook.py and sitecustomize.py, pin 1.106.2, then rotate every secret from a clean machine.
The code was clean. The release button wasn't. Your supply chain is only as safe as the laptop of whoever can press it.
On 22 October 2026 Railway said an unauthorized third party may have accessed encrypted environment variables and, for some, obtained the decrypted contents. It paused all deployments, told customers to rotate every secret and check logs, and said it will contact accounts confirmed as compromised. Railway has not said how many users are affected.
What to do
Rotate every secret stored on Railway today, including database passwords, API keys and private registry credentials. Then read your service logs for use of those keys you do not recognize.
A secret you only rotate after the breach was never really a secret. It was a countdown.
GitLab released 19.5.1 and backports on 22 October 2026 for CVE-2026-110264, an integer overflow in its regular expression compiler. GitLab says an unauthenticated user could run arbitrary code on the server under certain conditions. BleepingComputer does not report exploitation and says GitLab.com already runs the fix.
What to do
If you self-manage GitLab, upgrade today to 19.5.1, 19.4.3, 19.3.6, 19.2.9, 19.1.9, 19.0.10 or 18.11.13, whichever matches your line.
Two maximum-severity warnings in ten days. Patch day is now every week.
Microsoft published guidance on 22 October 2026 for apps that broke after Exchange Web Services was blocked by default in Exchange Online on 1 October. Admins can switch EWS back on with Set-OrganizationConfig -EWSEnabled $true, which Microsoft says takes effect within 4 hours. EWS shuts down for good on 1 April 2027, per BleepingComputer.
What to do
Find every integration that calls EWS and move it to Microsoft Graph, which Microsoft recommends. Until then, ask affected customers' admins to re-enable EWS.
A temporary switch has an expiry date. Build for the date, not the switch.
On 22 October 2026 Arista patched CVE-2026-114625, a SQL injection in on-premises VeloCloud Orchestrator that it says is under active attack. BleepingComputer reports an authenticated user can gain operator privileges and appear as a local superuser. Arista lists two attacker IP addresses and, per BleepingComputer, this is its fourth exploited zero-day this year.
What to do
If you run VeloCloud Orchestrator on-prem, install Arista's fixed build today and limit its web interface to admin networks. Review recent admin activity and block 74.82.197.86 and 38.54.50.151.
Four exploited zero-days in one year is not bad luck. It's a pattern you should plan around.
On 22 October 2026 CISA, the Canadian Centre for Cyber Security and partners published guidance on managing AI token consumption. It lists rising and unpredictable costs, reduced availability, degraded performance, token exhaustion, and denial-of-tokens or denial-of-wallet attacks as risks.
What to do
Set spend caps and alerts on every AI provider key, monitor token use over time, and cache reusable context. Write down which features break if your provider slows you down.
Your AI provider's meter is part of your threat model now. Watch it like you watch your logs.
On 21 October 2026 PHP released 8.5.12, 8.4.27, 8.3.36 and 8.2.35 to fix PHP2Shell, two critical flaws (CVE-2026-109618 and CVE-2026-109619) that allow unauthenticated remote code execution; SecurityWeek lists both at CVSS 9.2. Cloudflare, Patchstack, Hadrian and watchTowr reported exploitation attempts the same day, and watchTowr says none it saw has succeeded yet.
What to do
Upgrade to 8.5.12, 8.4.27, 8.3.36 or 8.2.35 today, rebuild your container images and restart every PHP worker. On PHP 8.1 or older, get a backport from your distribution or LTS vendor, or move to a supported release.
The warning gave you one day. The attackers used that day better.
Private keys for eight intermediate certificates that the Indian CA eMudhra issued in April 2025 could be recovered almost instantly; Mozilla's Dennis Jackson says they were effectively all zeroes apart from the last byte. Mozilla and Google blocked them on 17 October 2026, Apple on 19 October and Microsoft on 20 October, and eMudhra revoked the certificates.
What to do
Update browsers, operating systems and every app or container image that bundles its own CA list today. If you pin certificates, confirm none of them chains to an eMudhra intermediate issued in April 2025.
Encryption is only as strong as the dice someone rolled once. These dice had one face.
On 20 October 2026 DHS published an order in the Federal Register that tells federal civilian agencies to exclude DeepSeek products from procurement and remove them from their IT systems, or request an exception within 60 days of the order's issuance. It covers DeepSeek's affiliates and third-party products and services built using DeepSeek technology, and it is not a general ban on private or consumer use.
What to do
If you sell to US federal civilian agencies, list every model your product calls, including the ones inside vendor APIs, and replace any DeepSeek model before the agency asks.
Model choice used to be an engineering decision. In federal sales it is now a compliance line item.
On 20 October 2026 PHP warned of two critical remote code execution flaws, CVE-2026-109618 and CVE-2026-109619, and scheduled emergency updates for 21 October at 14:00 UTC. PHP rates each 9.2 on CVSS v4.0 but has not named affected versions or fixed releases; researchers at watchTowr say they have seen no exploitation.
What to do
Find every PHP host and container you run today. Keep a window open after 14:00 UTC on 21 October and install the update the moment PHP names the fixed releases.
A patch date is a starting gun for two sides. Be at the line before the other side is.
Attackers hijacked a developer's GitHub account and tampered with OVHcloud CLI releases v0.11.0 to v0.15.0. The compromised release appeared at 14:47 on 19 October and was blocked at 16:21 after a customer report. OVHcloud says v0.10.0 and v0.16.0 are clean, and source builds and Docker images were not affected.
What to do
If anyone ran OVHcloud CLI v0.11.0 to v0.15.0, treat every secret on that machine as compromised: clean it up and revoke the keys. Then turn on GitHub's release immutability for your own repositories.
A release you did not build yourself is a stranger's binary wearing your logo. Pin it, hash it, and lock your releases.
Zeabur says an attacker used leaked internal credentials on 14 to 15 October 2026 to enter its shared cluster. BleepingComputer reports over 3.2 million environment-variable records and 321,810 accounts were exposed. Zeabur is winding down: renewals stop and shared-cluster services pause on 1 December, and everything ends by 31 December.
What to do
If you host anything on Zeabur, rotate every credential stored there, revoke active sessions and rebuild any compromised servers. Start the move now: no extra migration time has been confirmed.
The host that stores your secrets is also your single point of failure. Know your exit before you need one.
TeamUpdraft found unauthorised access to its website over the weekend of 17 to 18 October 2026 and believes its customer database, including UpdraftCentral connection keys, was reached. Patchstack and Nethemba report attacks from 18 October, 15:32 UTC, adding administrators and at least 99 disguised backdoor plugins. UpdraftPlus plugin files are unaffected.
What to do
Update UpdraftPlus to 1.26.13 (2.26.13 Premium) or later, then delete the UpdraftCentral connection keys under Advanced Tools. Review admin users and plugins for strangers; if compromised, restore a clean backup.
The key to your backup tool is a key to the house. Rotating it is your job, even when the vendor was the one breached.
Rank Math PRO 3.0.123, downloaded from the vendor's site between 15 October 20:02 UTC and 16 October 07:12 UTC, carried malware. Once activated it set up persistent access, contacted rank1cdn[.]com and, per the vendor, exposed database credentials and API keys. The clean release is 3.0.124; only PRO was hit, not the WordPress.org version.
What to do
Search for admin users starting with wpsupport_, the cron marker #ms6v46 and the database entry _np5pvf754c. If PRO 3.0.123 was ever activated, restore a pre-15 October backup, block rank1cdn[.]com and rotate all passwords, keys and salts.
The update button is the most trusted button on your site. That is exactly why attackers want to sit behind it.
OpenCode said API keys that 1.4% of its Console users had stored for Anthropic, Google and OpenAI were exposed in an attack it calls ongoing. TechCrunch reports unauthorized charges on users' accounts, a few above $1,000. Anthropic and Google blocked keys they believed were exposed, OpenAI worked on refunds, and the open source app and CLI were not affected.
What to do
If you ever saved provider keys in OpenCode Console, revoke and replace them at each provider today. Check usage since 17 October and set a monthly spend cap on every AI key you own.
Every place that stores your API key is a place that can spend your money. Keep that list short, and keep it written down.
On 19 October 2026 GitHub said that during periods of heavy traffic it may temporarily require authentication for some clones and fetches of public repositories over HTTPS on github.com. SSH, pushes, authenticated HTTPS and other web and API traffic are not affected. GitHub says this is not a general authentication requirement for public repositories.
What to do
Search your CI for unauthenticated HTTPS clones of github.com. Use GITHUB_TOKEN in Actions and a GitHub App installation token in other CI systems, and add a retry where a job cannot authenticate.
Anonymous access is a courtesy, never a contract. Build as if the courtesy ends on your release day.
Microsoft says Windows Server 2019 and later, after the October 2026 security updates, can break Exchange: Outlook on the web sign-in fails, mail stays in the Outbox and mailbox replication struggles. The update capped Endpoint Mapper lookups at 512 simultaneous calls, which may be too few under heavy load. A fix is in progress; Microsoft advises against uninstalling the updates.
What to do
If you run Exchange on Windows Server 2019 or later patched in October, apply Microsoft's EpmapSavedContextCap registry workaround from an administrator command prompt, reboot, and keep the security updates installed.
Security patches change limits nobody wrote down. Watch your mail queue on patch week the way you watch a deploy.
Lior Aharonov · my take