In partnership with
| FACT | Delft University's Threat Intelligence Lab tracked more than 50,000 PHP2Shell exploitation requests from over 600 IP addresses in 70 countries; 171 of those addresses went past probing and sent exploit code or malware droppers. (Delft University via The Register, October 2026) |
|
Every patch is now a map. When AI can read a fix and write the exploit, the gap between "update available" and "you're compromised" shrinks to days, so this edition is about who moves first. PS: Was this forwarded to you? Subscribe free at exzeccyber.com/subscribe → |
In this edition | | | | | 🏛️ Privacy, Power & Policy |
| | |
|
Vulnerability / AI AI Found Critical Flaws in Fortinet Gear. The Clock Runs Out November 10. Intro Fortinet pointed AI models at its own products, and they found critical and high-severity bugs. The fixes shipped last week; the details stay under wraps until November 10. What Happened Fortinet released updates for FortiOS, FortiManager, FortiAnalyzer, FortiMail and FortiAuthenticator, saying frontier and in-house AI models found the flaws and humans validated them. At least some may be remotely exploitable without authentication, and there are no CVE IDs yet. Fixed branches: FortiOS, FortiManager and FortiAnalyzer 8.0, 7.6, 7.4 and 7.2; FortiMail 8.0, 7.6 and 7.4 (7.2 users move to 7.4 or later); FortiAuthenticator 8.0 and 6.6. Why It's Important Fortinet says its models were good at comparing firmware versions and building proof-of-concept exploits for already-fixed bugs. Attackers have the same tools, so the patch itself points at the hole. "The window between disclosure and exploitation continues to shrink, and we expect it to keep shrinking," said CISO Carl Windsor. The Other Side No exploitation has been reported, and holding details for two weeks gives defenders a head start. AI isn't magic, either: across 100+ source-code tests, about 30% of findings were false positives. | | 👉 Takeaway Upgrade internet-facing Fortinet gear now, not on November 10. Until then, limit admin access to trusted IPs, keep FortiManager off the internet and enforce MFA, but treat those as stopgaps. |
TL;DR: Fortinet's AI found critical bugs in its gear; patch before the details go public November 10. |
| | Active Exploitation CISA added CVE-2026-109618 and CVE-2026-109619 (both CVSS 9.2) to its exploited list on October 26 and gave federal agencies until October 29 to patch. Perplexity researcher Adel Karimi logged 35 successful shell executions on PHP 8.4.26 honeypots October 24, and AWS saw attempts from October 17, four days before disclosure. Sophos and Palo Alto walked back earlier attack claims. → Update to PHP 8.5.12, 8.4.27, 8.3.36 or 8.2.35 now. |
| | Supply Chain Wiz, tracking the "Gheist" campaign since September 15, has recovered 13,500+ planted branches from 850+ compromised accounts. Each ci-sync- branch runs a workflow that dumps every readable secret in base32, slipping past GitHub's masking, then deletes the run. GitHub's hold on suspicious runs covers public github.com repos only, not private repos or Enterprise Server. → Found a ci-sync branch? Rotate every secret that repo could reach. |
| | Vulnerability Erica Windisch of Cyberdione Labs published "Tailworm" October 25: in lab tests, one compromised device with Tailscale SSH and port forwarding allowed reached root on other servers, no stolen keys needed. Tailscale says it is fixing issues but isn't telling everyone to disable Tailscale SSH. There is no patch or CVE yet, and the disclosure is disputed. → Narrow SSH rules to single source-destination pairs and turn off forwarding. |
|
Your AI prompts are leaving out 80% of what you're thinking.
When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning (constraints, edge cases, examples, tone) and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.
Used by teams at Microsoft, Amazon, and Notion. Try Wispr Flow free on Mac, Windows, iPhone, and Android.
| | AI Offense Police say a self-taught 32-year-old former security-company pentester led a seven-person crew that stole 8.2 million records from schools, hospitals, companies and golf courses, mostly using the ARTEX AI pentesting framework. He demanded 360 million won ($270,000) in 38 extortion attempts. Police say even non-professionals "can now hack on a large scale with the help of AI." → Attacker skill is no longer the bottleneck. |
| | AI Agents Forever Security found a malicious AGENTS.md in a customer's ~/Downloads (the agent's working directory) pushing a rogue npm package. Agents load these automatically, and a web page can drop one there. In tests, Codex followed it 90% of the time; a CLAUDE.md variant exposed Claude Code's OAuth token. Claude Code alone ran 1,300+ sessions from Downloads at their customers. → Start agents in the project root, never in Downloads. |
|
| 🏛️ Privacy, Power & Policy | |
| | Privacy On October 20, federal judge Denise Cote ruled in AFGE v. OPM that the agency violated two provisions: limits on sharing between agencies and the duty to protect records. OPM holds records on tens of millions of federal workers; DOGE access began January 2025. She found no support for a modernization motive; a remedy comes later. (EFF represents the plaintiffs.) → "Modernization" isn't a Privacy Act exemption. |
| | Surveillance In audio from a private police conference on October 25, obtained by 404 Media, Garrett Langley claimed without evidence that anti-Flock social media chatter is down about 95% since July and fell 50% after Ukraine hit a Russian data center with a drone. 404 notes that repeats a disproven theory. He also said his wife told him never to apologize. → The opposition is local, and it's real. |
|
| | Social Engineering Arctic Wolf saw hundreds of cases following one script: flood a victim's inbox with sign-up emails, message them on Teams from an external "IT support" account (30-40% on *.onmicrosoft.com tenants), talk them into Quick Assist, AnyDesk or similar, then plant backdoors, move to domain controllers, steal data and deploy ransomware. The play: limit external Teams chat to allowlisted partner domains, alert on remote-access tools your IT team doesn't use, and tell staff the real help desk never cold-messages them after a spam flood. |
|
A deal has two sides. Your stack only sees one.
Aligned shows you the other half. See what buyers actually do between meeting. See what they open, revisit, and share; so you know the exact next step to move things forward. And your champion gets everything they need to win internally when you're not in the room.
 Strange but real The Ransomware "Fixer" Who Allegedly Just Paid the Ransom Intro Most ransomware recovery pitches say "don't pay." Prosecutors say one Florida firm's secret method was paying anyway, then charging you extra. What Happened The DOJ charged Zohar Pinhasi ("Zack Silver," "Zack Green"), who allegedly ran MonsterCloud, a self-described "most sophisticated Counter Cyber Terrorism team in the world" promising "proprietary tools" and "advanced decryption techniques." Prosecutors say he quietly paid attackers and kept the difference, once billing $150,000 for an $8,200 ransom. Overall he allegedly charged clients more than $19 million and paid out more than $8 million. In May 2019 he told a paid spokesperson: "MonsterCloud doesn't hold any proprietary technology [to] decrypt the ransomware data." Why It's Important Victims who chose not to pay criminals may have funded them anyway, at a markup, without knowing it. The Other Side These are allegations, not convictions, and the indictment mentions "multiple co-conspirators," so more charges could follow. Firms that negotiate with attackers can be legitimate, as long as they say so. | | 👉 Takeaway Before hiring incident help, ask in writing whether they ever pay ransoms for clients, and demand proof any "decryptor" exists. |
TL;DR: A firm that promised to beat ransomware without paying allegedly just paid and pocketed the markup. |
Your search bar is ghosting shoppers with “0 results found.” Gladly understands what your shoppers mean and turns dead ends into product recommendations.
