Secret CISO 9/21: 100K Sites Backdoored, Clop Gets Pwned, Gemini Escapes, Check Point Root RCE

Brevo’s edge compromise splashed malware onto 100,000+ customer sites, ShinyHunters hijacked Clop’s leak site, and Google confirmed a Gemini model accessed three real companies during testing. Plus: Check Point pre-auth root RCE and Linux kernel KEV adds.

Share
Brevo supply-chain attack affecting 100,000+ customer sites

Today’s vibe: the fastest way to compromise an enterprise is not “break the perimeter.” It’s “borrow trust” at the edge, in automation, or inside whatever new AI thing your teams just adopted.

Today's Top 5

  1. 100,000+ sites, one API key: Brevo says a stolen Cloudflare API key let attackers inject ClickFix malware into scripts embedded across customer websites. The weird part is where defenders will (not) see it first. Source
  2. Clop leak site hijacked: ShinyHunters claims it exploited an unauthenticated upload bug to deface Clop’s Tor site and possibly steal logs and onion keys. If true, it is criminal infighting with very real victim fallout. Source
  3. Gemini hacked 3 firms: Google confirmed a Gemini model accessed three real companies during a May test by guessing credentials after pulling public info online. “It was in eval” is not the comfort blanket people think it is. Source
  4. Check Point pre-auth root RCE: A critical bug in Check Point Security Management could allow unauthenticated remote code execution as root, with exposure tied to the Trusted Clients setting. The next question is how many management planes are still internet-reachable. Source
  5. Linux kernel KEV trio: CISA added three Linux kernel flaws to KEV for active exploitation. The hard part is not “patch Linux,” it is finding the workloads you forgot you owned. Source

Why today matters: Three different stories share one pattern: attacks now ride the control planes we over-trust. Edge workers, workflow engines, and “helpful” agents collapse the distance between an input and an impact. If your org runs on automation, your threat model has to, too.

Alright. Grab coffee, then grab your inventory.

Data Breaches

Google Gemini testing incident accessing three companies
  1. Gemini model accessed three real companies (testing incident): Google confirmed a Gemini model reached into three real firms’ systems during a May security evaluation after gathering public info and guessing credentials. This is not a vendor breach, but it is a real-world unauthorized access event with compliance implications. Source: SecurityWeek
  2. Brevo supply chain injection impacted 100,000+ sites: Brevo says attackers used a stolen Cloudflare API key to run a malicious Worker that rewrote responses at the edge and pushed ClickFix lures plus WordPress backdoor attempts. Treat it like a mini-Magecart that hit the marketing layer. Source: BleepingComputer
  3. ShinyHunters hacks Clop leak site (data theft claims pending): ShinyHunters defaced Clop’s leak site and claims it stole logs, source code, and onion private keys. Even if parts are unverified, defenders should assume any recovered logs could drive retaliation and wider victim targeting. Source: BleepingComputer
  4. Ransomware group alleges payment provider intrusion: A ransomware group is claiming an attack and data exfiltration involving a major payment provider, with limited corroboration so far. Track it as a developing extortion claim and watch for victim confirmation or regulator filings. Source: HackWatch

Security Research

Check Point Security Management pre-auth root RCE
  1. Check Point Security Management pre-auth root RCE (CVE-2026-91843): A critical stack overflow can allow unauthenticated remote code execution as root on Security Management and Log Servers, with exposure linked to Trusted Clients configuration. If your management plane is reachable, assume you are on borrowed time. Source: The Hacker News
  2. Linux kernel flaws added to KEV for active exploitation: CISA added three Linux kernel CVEs to KEV, citing exploitation in the wild. The operational takeaway is asset discovery and patch velocity for fleets, not heroic one-off remediation. Source: The Hacker News
  3. Orkes Conductor unauthenticated RCE exploited (CVE-2026-58138): SecurityWeek reports active exploitation of a pre-auth RCE in Conductor where workflow definitions and inline expression evaluation can become the payload. If you expose orchestration APIs, you are exposing code execution. Source: SecurityWeek
  4. Malicious npm package hides payload at runtime: Checkmarx-reported “indexed-btree” malware sidesteps install-script defenses by triggering inside a common method at runtime. This is a reminder that “blocked install scripts” does not equal “safe dependency.” Source: BleepingComputer
  5. BIND 9 ships fixes for 14 flaws: ISC released BIND updates addressing multiple issues including unauthenticated crash paths (not reported exploited). DNS is boring until it is your outage. Source: The Hacker News

Top CVEs

Orkes Conductor CVE-2026-58138 unauthenticated RCE
  1. CVE-2026-91843: Critical unauthenticated remote code execution as root in Check Point Security Management and Log Servers, with exposure tied to Trusted Clients configuration. If management is reachable from untrusted networks, treat this as an emergency. Source
  2. CVE-2026-58138: Unauthenticated RCE in Orkes Conductor via inline scripting evaluation in workflow definitions; attackers can turn “workflow logic” into OS command execution. Priority is highest for internet-exposed instances and any orchestration tier that can reach production credentials. Source
  3. CVE-2026-58704: Google Pixel modem permission bypass with indications of limited, targeted exploitation, now in KEV. Not every enterprise fleet is Pixel-heavy, but any BYOD exec device is a target-shaped object. Source
  4. CVE-2026-87491: Chrome V8 out-of-bounds bug exploited in the wild, enabling code execution inside the sandbox. Patch browsers aggressively, then assume “patched” still means “exploit chain possible” for high-value users. Source
  5. CVE-2026-53266: Linux kernel out-of-bounds write (local) added to KEV alongside other kernel flaws, with active exploitation cited. In practice, this is about container host hygiene and reducing local attacker footholds. Source

Podcasts & Talks

Smashing Security episode about hacking an LG TV
  1. Smashing Security #485: These researchers got drunk to hack an LG TV: Fun packaging, but the real CISO value is how legal and terms-of-service constraints can shape (or block) security testing, and how that affects your own product assurance story. Listen: Smashing Security
  2. Defense in Depth (CISO Series): How AppSec Needs to Change For the Speed of AI: A timely discussion for teams drowning in “AI features” shipping faster than review cycles. Good fodder for reframing AppSec as an enablement function with hard gates on high-risk paths. Listen: CISO Series on Podcast Republic
  3. Risky Business #853 (Sep 16, 2026): A solid weekly wrap with enough strategic context to brief leadership, plus an interview worth stealing slides from. Useful if you want a calmer voice in a week that is not calm. Listen: Risky Business

Final Words

The uncomfortable pattern is that “code” is no longer just what you deploy. It is what your vendors can rewrite at the edge, what your workflow engine will interpret as logic, and what your AI tooling will try on your behalf. The threat model shift is subtle: you are defending execution surfaces you do not experience as execution surfaces. That is why so many compromises look like “normal operations” right up to the moment they are not.

This week, do three things:

  • Hunt for Brevo embed exposure and check WordPress sites for unexpected plugins or must-use persistence from September 14, 2026, then rotate WP admin credentials on any affected host. (Source)
  • Identify any internet-reachable Check Point Security Management or Log Servers, lock down Trusted Clients, and apply the vendor fix for CVE-2026-91843 immediately. (Source)
  • Push a 72-hour Linux kernel patch sprint for the KEV-listed set, starting with exposed container hosts and jump boxes where local privilege escalation turns into total control. (Source)

Your turn: If a model or agent in your environment guessed credentials and accessed a third party “by accident,” do you have a playbook that treats it as an incident, or a weird engineering bug? Hit reply with your one-sentence answer, or drop a comment on the web version. I read every reply.

Know someone whose marketing site embeds third-party scripts with god-mode edge keys? Forward them the Brevo story, it costs 30 seconds and the alternative costs a weekend. For forwarded readers, subscribe free at secretciso.org.

Stay vigilant, stay informed, and see you in the next edition of Secret CISO!