Weekly security intelligence digest covering the most critical vulnerabilities, threats, and breach news from the past week.

🚨 Critical: CISA Known Exploited Vulnerabilities

These vulnerabilities are being actively exploited in the wild. Immediate action required.

CVE-2026-45659: Microsoft SharePoint Server Deserialization of Untrusted Data Vulnerability

Vendor/Product: Microsoft SharePoint Server

Description: Microsoft SharePoint Server contains a deserialization of untrusted data vulnerability which allows an authorized attacker to execute code over a network.

Required Action: Apply mitigations in accordance with vendor instructions, ensuring compliance with CISA’s BOD 26-04 Prioritizing Security Updates Based on Risk (see URL in Notes) guidance and CISA’s “Forensics Triage Requirements” (see URL in Notes). Follow applicable BOD 26-04 guidance for cloud services or discontinue use of the product if mitigations are unavailable. Stakeholders are responsible for evaluating each asset’s internet exposure and ensuring adherence to BOD 26-04 patching guidelines.

CISA Due Date: 2026-07-04

Reference: CVE-2026-45659 - NVD

Expert take: The story here isn’t the bug, it’s that Microsoft called it wrong. When the patch shipped in May, Microsoft rated this “Exploitation Less Likely.” Six weeks later CISA put it on the KEV list because it’s being exploited for real. So a “probably fine” flaw became a federal 3-day emergency. That’s worth sitting with if your patch process leans on vendor severity ratings to decide what waits.

A couple of quirks made this messier. The fix actually went out in the May updates, but Microsoft forgot to include the CVE in the release notes and only disclosed it on May 21. So some teams patched without knowing, and others never flagged it as important. Either way, “we ran May’s updates” is not the same as “we’re covered here” — go verify the specific build.

On the bug itself: it’s a deserialization flaw, meaning SharePoint takes serialized data from a request and rebuilds it into objects without checking what it’s rebuilding. Feed it a crafted payload and you run code on the server. The one thing keeping this at CVSS 8.8 instead of a perfect 10 is that the attacker needs to be authenticated — but only barely. Site Member permissions are enough, no admin rights needed. In a big org with lots of low-level SharePoint accounts, that’s a low bar.

Why care so much about SharePoint specifically: it’s where companies keep their documents, intranets, and internal collaboration, and it’s usually reachable from the internet. It’s also the same platform that got hammered by the ToolShell mass-exploitation wave in 2025. So this isn’t a new problem, it’s a familiar target getting poked again.

What to do: confirm the May patch is actually on every SharePoint server (don’t trust “Windows Update said success” — on-prem SharePoint needs farm-aware validation). Then review who has Site Member or higher, especially guest and external accounts. And because attackers in past SharePoint campaigns stole ASP.NET machine keys to keep access even after patching, assume that if your server was exposed before you patched, the patch alone might not have kicked them out. Rotate keys and hunt backwards. Note this is on-prem only — SharePoint Online / M365 isn’t affected.

đź“° This Week’s Security News

JadePuffer ransomware used AI agent to automate entire attack

Researchers identified what they believe is the first documented case of a ransomware operation, JadePuffer, conducted entirely by a large language model (LLM) agent. […]…

Read more: JadePuffer ransomware used AI agent to automate entire attack

Expert take: This is the one everyone’s going to talk about, and the framing matters. Sysdig calls JadePuffer the first documented ransomware attack run end-to-end by an AI agent. No human clicking through the steps. The agent did recon, stole credentials, moved laterally, set up persistence, escalated privileges, encrypted data, and dropped the ransom note on its own.

The detail that stands out: it adapted on the fly. In one spot it went from a failed login to a working fix in 31 seconds. That’s the actual shift here — not that any single technique is new, but that the AI chained them together and problem-solved in real time, the way a human operator would, just faster.

Here’s the part people will miss in the hype, and it’s the important part. None of the techniques were sophisticated. The way in was CVE-2025-3248, an old unauthenticated RCE in Langflow — the same AI-workflow tool that showed up on CISA’s KEV list back in May. It’s been patched since April 2025. The attack then pivoted to a server running a years-old Nacos auth bypass from 2021 and a default signing key nobody changed. So the “AI superweapon” broke in through neglected, internet-facing junk that should’ve been patched years ago.

That’s the real lesson. AI doesn’t give attackers new powers so much as it makes the cheap, boring attacks free to run at scale. Spraying every known old vulnerability across the whole internet used to cost attacker time. Now an agent does it for the price of running the model — and if it’s running on stolen API keys, close to free. So the long tail of unpatched, forgotten systems just got a lot more dangerous.

Good news buried in the report: the AI-written payloads had natural-language comments explaining their own reasoning, which is weird and unusual for malware and gives defenders a new thing to detect on. But don’t lean on that. The actual fix is boring and unchanged: patch your internet-facing stuff, kill exposed admin interfaces, change default keys. JadePuffer didn’t beat good hygiene. It found people who didn’t have it.

Cisco finally confirms attackers exploiting Unified CM flaw

Cisco confirmed that attackers are now exploiting a Unified Communications Manager (Unified CM) vulnerability patched in early June. […]…

Read more: Cisco finally confirms attackers exploiting Unified CM flaw

Expert take: Note the word “finally” in the headline. That’s the whole story. Cisco patched this (CVE-2026-20230) on June 3 and said no sign of exploitation. Defused spotted real attacks around June 22. CISA added it to KEV on June 25 with a 3-day deadline. Cisco didn’t update its own advisory to confirm exploitation until July 1. So there was a stretch where outside researchers and CISA were saying “this is being attacked” and the vendor’s page still said otherwise.

The bug is an SSRF (server-side request forgery) — you send a crafted HTTP request and trick the server into doing things it shouldn’t. In this case it lets an unauthenticated attacker write files to the OS, and from there work up to root. Cisco actually rated it Critical even though the CVSS is 8.6, specifically because the endgame is root. Unified CM runs enterprise phone and collaboration systems, so root on it is a deep foothold in your comms infrastructure.

One thing that genuinely lowers the risk: it only works if the WebDialer service is enabled, and that’s off by default. So a lot of orgs aren’t exposed at all. But the early exploit activity was file-write probes dropping test files to find vulnerable boxes, which means attackers are actively mapping who’s open. Shadowserver counts 200+ Unified CM instances sitting on the internet.

What to do: patch to 14SU6 (15SU5 isn’t out until September). If you can’t patch right now, just disable WebDialer — if you’re not using it, there’s no reason it should be on. And the meta-lesson: don’t wait for the vendor to admit exploitation before you act. KEV listing and third-party telemetry were flashing red a week before Cisco’s advisory caught up. Your patch priority should follow the attacks, not the vendor’s comfort level.

âś… What You Should Do This Week

  • Immediate: Patch CVE-2026-45659 (actively exploited)
  • Verify: Check your systems against CISA KEV catalog
  • Monitor: Review Azure AD sign-in logs for suspicious activity
  • Audit: Verify MFA is enforced for all privileged accounts
  • Backup: Test your disaster recovery procedures

📬 Stay Updated

Subscribe to receive weekly security digests directly in your inbox.

Questions or feedback? Contact us

GRC Vitrix provides cloud security and compliance intelligence for financial services professionals. This digest is curated from publicly available sources including CISA, Microsoft MSRC, and industry news.