This Week in Security: Firewall Fixes, a Rented Banking Trojan, a Blank PDF and Stolen Tokens. What to Actually Do

By Thorsten Meyer | October 29, 2026 | ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Every Thursday, The Hacker News publishes a long digest of the week’s security news. This week’s has thirty stories (The Hacker News, October 29). Nobody needs thirty. I picked the six that matter most for businesses and for ordinary users, went back to the original advisory or research report for each one, and boiled each down to a short story and one clear action.

Two of the six are about your company’s front door. Two are about your phone. One is about your inbox. And one is about the list that security teams use to decide what to fix first.

A word on sources before we start. Most numbers below come from the security companies that found the problem. They are good at this work, and they also sell protection against it. Where a number is a company’s own count, I say so.

Key takeaways

• Firewalls first. Fortinet shipped critical and high-severity fixes for FortiOS, FortiManager and FortiAnalyzer on October 22 and asks customers to treat them as a priority change. The details stay private until November 10. At the same time, the FBI and the US Secret Service warn that the “FortiBleed” password-theft campaign against Fortinet firewalls is still active (Fortinet; FBI and USSS).
• Bank fraud now comes as a subscription. CloudSEK found an Android crime kit called MR ROBOT that criminals rent, complete with a licence countdown, to build fake bank apps that steal one-time codes and secretly forward calls (CloudSEK, October 28).
• The dangerous invoice was blank. Trellix traced a hidden backdoor to an empty PDF disguised as a FedEx tax invoice. It only worked because the victim clicked “yes” on the PDF reader’s warning (Trellix, October 26).
• An app with zero permissions got work tokens. Researchers showed that Microsoft Authenticator on Android could be tricked into handing another app the login tokens of a work account. Microsoft has fixed it (Enclave, October 23; Microsoft, CVE-2026-80097).
• “Hi, I’m ChatGPT” is not an ID card. Cequence tracked one scanner that rotated through about 32 fake AI-bot identities while asking websites for password files and cloud keys (Cequence, October 28).
• The official must-patch list is a floor, not a ceiling. CISA gave US federal agencies three days to fix PHP2Shell. A critical flaw in the document tool Gotenberg, with 1,358 recorded attack attempts in 30 days, is not on CISA’s list at all (CISA KEV; VulnCheck, October 26).

Products Worth Considering

Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok

AmazonView Latest Price

Amazon Kindle Paperwhite Signature Edition (12th Generation, 2024), 32 GB, Metallic Black

AmazonView Latest Price

Bose TV Speaker (2nd Gen) Compact Soundbar with Bluetooth and HDMI

AmazonView Latest Price

As an affiliate, we earn on qualifying purchases.

The short answer

None of this week’s attacks needed magic. Each one needed one of four things: a login page that faces the internet, a password that was stolen months ago, a person who clicks “Allow”, or a piece of software that believes what another program says about itself.

That is good news, because all four have boring, well-known fixes. Below, each story ends with one action. If you only have ten minutes, do the actions and skip the stories. The patch order for IT teams and a checklist for everyone else are further down.

1. The firewall’s own backup file

A firewall is the guarded front door of a company network. Fortinet’s FortiGate is one of the most common. This summer, researchers found that a criminal group had left its own server open. On it sat a complete assembly line for breaking into FortiGate firewalls. The campaign got a name: FortiBleed.

On October 6, the FBI and the US Secret Service warned that it is still running (joint advisory). The agencies cite the security firm SOCRadar, which verified more than 86,644 compromised devices across 194 countries. The method is not a clever new bug. The attackers reuse passwords leaked in earlier incidents, guess weak ones, and crack password files stored with an old method (legacy SHA-256) on a rented graphics-card cluster. Then they create new administrator accounts, sometimes lock the real IT team out, and sell the access. The agencies name the ransomware groups INC/Lynx and Payload as buyers.

On October 27, Hudson Rock added a second batch (Hudson Rock, “FortiBleed 2.0”). It analyzed 1,703 firewall configuration reports from 1,689 devices and counted 40,573 records with passwords in plain text. Among them: 37,076 local user accounts and 1,576 records of the account that connects the firewall to the company directory (Active Directory). A firewall backup is the building’s master key ring and the staff list in one envelope. Hudson Rock itself stresses that these are records, not people, and not all of them still work: 7,437 of the local accounts were marked as disabled.

Fortinet says FortiBleed “is not a new Fortinet vulnerability”. It points to credentials reused from earlier incidents and to password guessing against devices with weak passwords and no second factor (Fortinet PSIRT, June 19).

Separately, and this is the “critical firewall flaws” part of the headline, Fortinet released coordinated security updates for FortiOS, FortiManager and FortiAnalyzer on October 22, after earlier updates for FortiMail and FortiAuthenticator (Fortinet, October 21). They include critical and high-severity fixes found with AI-assisted testing; Fortinet thanks Anthropic, OpenAI, Google and SpaceX for help. The public advisories follow on November 10. Fortinet’s own reasoning for the rush is worth reading: AI tools can now help attackers “reverse-engineer patches and build working exploits in hours or days, rather than weeks”. We looked at that shrinking window in Anthropic’s AI Built Five Working Attack Chains.

For context: Fortinet products account for nine of the entries CISA added to its list of actively exploited flaws in 2026, the latest on October 16 (my count of the catalog). Fortinet is not alone: network edge boxes are attackers’ favorite door, as our piece on Cisco SD-WAN malware shows.

The one action: If you run Fortinet, install the October 22 releases, take the admin login off the internet, and reset every VPN, administrator and directory-connection password. Then look for admin accounts nobody on your team created. The FBI’s ladder for the admin login: allow only trusted addresses (good), a stricter local policy (better), no admin access from the internet at all (best).

2. Bank fraud with a licence countdown

One screenshot in CloudSEK’s report shows a status line: “Licence Active — 26 days remaining.” It is not an antivirus subscription. It is a criminal’s subscription to a banking-trojan kit.

CloudSEK calls the kit MR ROBOT (CloudSEK, October 28). It targets customers of Indian banks. A master operator rents access to “affiliates”, who use a Telegram bot called “APK Shield” and a web control panel (version 7.0) to build personalized fake apps on demand. The catalog has about 36 lures: bank apps, gas, water and electricity bills, phone top-ups and fake traffic fines. Victims get a link that leads to a fake app-store page.

Once installed, the fake app reads incoming text messages, including one-time codes. It silently switches on call forwarding, so the bank’s fraud-check call rings at a criminal’s phone instead. And it shows fake screens that ask for the card number, PIN and the PIN for India’s UPI payment system. To avoid detection it waits 48 hours or three launches before acting and hides its icon. One affiliate’s exposed database held 153 infected devices and 5,407 intercepted text messages, with call forwarding active on 58% of the victims.

The numbers are from one affiliate and one country. The business model is the bigger story: when crime kits can be rented, you do not need skill to run a bank-fraud campaign, only a subscription. CloudSEK links the techniques to earlier kits in the same region.

The one action (for users): Never install a bank app from a link in a text message, WhatsApp or Telegram. Banks do not distribute apps that way. CloudSEK also advises keeping Google Play Protect on, checking which apps may read your text messages and make calls, and dialling ##21# to cancel call forwarding (the code can vary by carrier; your carrier can confirm). For banks: CloudSEK recommends moving from text-message codes to app-based or FIDO security keys.

3. The blank invoice

Trellix’s incident team was called in because a PDF had done something PDFs should not do: start a program. The trail led back to an email in Korean: “[FedEx] A tax invoice has been issued”, 217,740 won, with an invoice number and a due date (Trellix, October 26).

The attachment was a ZIP file with a PDF inside. Open it and you see a blank page. Behind the page, the file tells the PDF reader to launch a hidden command as soon as it opens. There is one catch for the attacker: the PDF reader asks first. Per Trellix, the attack “relies on social engineering to trick the user into approving the PDF reader’s security confirmation prompt.”

After that click, everything runs in memory, without a file on the disk that antivirus could scan. The malware blinds the computer’s security sensors, hides inside normal Windows processes, and finally loads a backdoor that Trellix calls Sirius: remote control, screen capture, modules aimed at financial theft, and ways to survive even a reset of Windows. Trellix describes one customer case and gives no victim count, so we do not know how widespread this is.

The one action (for users): When a PDF asks permission to open or launch something, click no. A real invoice has content. A blank one that asks for permission is an alarm bell. Check invoices on the sender’s own website, not through attachments. For IT: Trellix recommends enabling the Microsoft Defender rule “Block Adobe Reader from creating child processes”, restricting PowerShell and logging what it runs, and alerting whenever a PDF program starts PowerShell.

4. The app that said “I’m Teams”

Researchers at Enclave wrote an Android app with zero permissions. It told Microsoft Authenticator: I am Teams, please give me a token for this work account. Authenticator checked something that sounded right, whether the request carried the correct return address for Teams. It never checked whether the app asking was actually Teams. In the researchers’ tests, the tokens they got gave “full delegated mailbox, files and calendar access” (Enclave, October 23).

Why tokens matter: two-factor login guards the door. A token is the wristband you get once you are inside, and nobody checks your ticket again. We covered the same idea for AI tools in Attackers Now Go After Your AI Tools.

Microsoft fixed the flaw as CVE-2026-80097 on September 8 and rates it “Important” with a score of 8.6 out of 10 (Microsoft). Microsoft’s description says the attacker must get a malicious app onto the phone and convince the user “to complete an authentication flow”. Enclave says no interaction was needed beyond Authenticator being installed with a work account signed in. Microsoft marks exploitation as “less likely” and listed it as not exploited at release. It is not on CISA’s list.

The one action: Update Microsoft Authenticator on Android. Enclave says versions before 6.2606.4246 are affected. For companies: require a minimum Authenticator version through your device management, and do not allow apps from outside the official store on phones that hold work accounts.

5. Thirty-two fake AI bots

Knock, knock. “Hi, I’m ChatGPT.” Cequence tracked one actor that posed as about 32 different AI bots, including crawlers and assistants from OpenAI, Anthropic, Meta and Perplexity (Cequence, October 28). It did not read web pages. It asked for about 90 file locations where secrets tend to live: configuration files with passwords (“.env” files), cloud credential files and private SSH keys.

Each address fired 50 to 190 requests in about eight seconds and moved on. The actor used more than 300 cloud addresses in one day and nearly 400 in a week, from at least late September to late October, against banks, retailers, telecoms, energy and software companies. On one day it reached 48% of the Cequence customers that saw any AI-bot traffic. At several companies, more than 80% of what was labelled “AI bot traffic” was this impostor. The giveaway: its fake Claude identities all carried the contact address of Anthropic’s real crawler, a copied-and-pasted string.

Why this matters: many websites now let AI crawlers in based on the name they announce. A name is not an ID. Cequence says nearly all of this actor’s requests ended in a block or an error at its customers; we do not know how many other sites handed something over.

The one action: Make sure your web server never serves .env, configuration or key files, and check the staging and admin sites too. Verify AI bots by their published address ranges or signed requests, not by the name in the request. Expect blocklists of addresses to go stale within days.

6. The must-patch list is a floor

CISA, the US cyber agency, keeps a catalog of flaws known to be exploited, the KEV list. US federal agencies must fix them by a due date, and many companies use it to decide what to patch first.

On October 26, CISA added the two PHP2Shell flaws in the PHP programming language (CVE-2026-109618 and CVE-2026-109619) with a due date of October 29. Three days. That is today. The web security firm Wordfence says that, as of October 28, it knows of no WordPress site compromised through PHP2Shell, but it recorded about 77,600 related attack requests from more than 2,000 addresses between October 23 and 27. Attempts are not successes. The fix is to update PHP to 8.5.12, 8.4.27, 8.3.36 or 8.2.35 and restart it; updating WordPress does not update PHP (Wordfence, October 28). Our PHP2Shell checklist has the details.

Now the other side. Gotenberg is an open-source tool that turns documents into PDFs; companies use it for invoices and reports. Its flaw CVE-2026-42589 is rated 9.8 out of 10 and was fixed in version 8.31.0 in May. VulnCheck recorded 1,358 exploitation events from 128 addresses in 30 days, from at least seven groups, from scanners to cryptominers to a custom backdoor (VulnCheck, October 26). It is not on CISA’s list; I checked the catalog version of October 28. VulnCheck put it on its own list on June 26.

For scale: CISA added eight entries between October 22 and 28 (my count), from PHP and the Brotli compression library to Adobe Commerce, the Linux kernel and VMware ESX.

The one action: Patch everything on the KEV list first. Then treat “exploited, according to a credible researcher” plus “reachable from the internet” as the same priority, even before CISA lists it.

What ties it together

Every story is about borrowed trust. The firewall trusts a password stolen months ago. Authenticator trusted what an app said about itself. Websites trust a bot’s name. The PDF reader trusts one click. And banks trust a text message that a trojan can read. Each action above removes one piece of blind trust.

What this means for companies

Here is the patch order I would use this week. It is my reading of the advisories above, not an official ranking.

1. Today: internet-facing edge devices. Install the Fortinet releases of October 22. Take admin logins off the internet. Reset VPN, admin and directory-connection passwords. Look for admin accounts nobody created.
2. Today: everything on CISA’s list whose due date has passed, including PHP2Shell. Update PHP and restart it.
3. This week: exploited software that is not on CISA’s list yet. If you run Gotenberg, move to 8.31.0 or later and check it is not reachable from the internet.
4. Identity and tokens. Require a current Microsoft Authenticator version on managed phones. Revoke sessions when a phone goes missing. The FBI recommends phishing-resistant multi-factor login for all remote access and admin accounts.
5. Email and documents. Turn on the Defender rule that stops PDF readers from starting other programs, and alert on PDF programs that start PowerShell.
6. Your website. No secrets in files the web server can reach. Verify bots by address range or signature, not by name.
7. Phones and customers. Block apps from outside the official store on phones with work accounts. If you are a bank, move customers away from text-message codes.

What this means for users

• Update. Your phone’s system, Microsoft Authenticator if you use it for work, and your banking apps.
• Bank apps only from the official store. Never from a link in a message, however urgent it sounds.
• Watch for missing calls. If calls stop reaching you or your bank says it called and you never heard it, check call forwarding.
• Say no to PDFs that ask permission. A document that wants to launch something is not a document.
• Check invoices at the source. Log in to the sender’s website instead of opening the attachment.
• Keep your bank password to yourself, also from AI agents. Bank of America recently reminded customers of exactly that. Our guide Don’t Give Your AI Agent Your Bank Password shows safer ways to let an agent handle money.

Honest limits

• Company numbers. CloudSEK, Trellix, Cequence, Hudson Rock, VulnCheck and Enclave are security companies. Their figures are their own counts, mostly from their customers, sensors or exposed criminal servers. I could not check them independently.
• Fortinet’s fixes are still a black box. We will not know what the October 22 releases fix until the advisories come out on November 10. Fortinet does not say whether any of these flaws are being exploited.
• FortiBleed figures have two sources. The 86,644 compromised devices are SOCRadar’s count as cited by the FBI and Secret Service. Hudson Rock’s 40,573 are records with plain-text secrets, not people and not working logins.
• Microsoft and the researchers disagree on whether the Authenticator attack needed the user to do anything.
• Two stories are regional. MR ROBOT targets India and the Sirius email targeted Korean speakers. The techniques travel; the numbers do not.
• What I left out, and why. The new infostealer that targets AI tools (we covered it this morning), the Bank of America warning (covered on October 28), Cohere’s report on the Qwen model (covered on October 27), a session-stealing remote-access tool known so far mainly from the seller’s own advertising, and several research reports that are important but do not come with one clear action for most readers.

Verdict

This was not a week of exotic attacks. It was a week of old weak spots with new tooling: AI that finds bugs faster on both sides, crime kits for rent, and impostors that wear a trusted name. The defence is the same list it has always been, done faster: patch what faces the internet, change passwords that may have leaked, stop trusting names and clicks, and keep your apps current. Six actions, most of them under an hour.

Sources

• The Hacker News, “ThreatsDay” bulletin (tip only), October 29, 2026: thehackernews.com
• Fortinet, security hardening release for FortiOS, FortiManager and FortiAnalyzer, October 21, 2026: fortinet.com
• Fortinet PSIRT, analysis of reported credential compromise of FortiGate devices, June 19, 2026: fortinet.com
• FBI and US Secret Service, joint advisory “FortiBleed Operations Continue Targeting Exposed Systems”, October 6, 2026: ic3.gov (PDF)
• Hudson Rock, “FortiBleed 2.0”, October 27, 2026: infostealers.com
• CloudSEK, “Inside MR ROBOT”, October 28, 2026: cloudsek.com
• Trellix, “Unveiling Sirius Backdoor”, October 26, 2026: trellix.com
• Enclave, “AuthMask”, October 23, 2026: enclave.ai
• Microsoft Security Response Center, CVE-2026-80097, September 8, 2026: msrc.microsoft.com
• Cequence, “Attackers Are Using AI Agent Impersonation to Hunt for Credentials”, October 28, 2026: cequence.ai
• CISA, Known Exploited Vulnerabilities catalog, version 2026.10.28: cisa.gov
• Wordfence, PHP2Shell update, October 28, 2026: wordfence.com
• VulnCheck, “Cache Me If You Can”, October 26, 2026: vulncheck.io

Researched and drafted with AI assistance, including Anthropic’s Claude.