Jul 19 · Industry News

You Patched SharePoint. Did You Get the Keys Back?

Today is the federal deadline to patch on-premises SharePoint against CVE-2026-58644 — a critical zero-day already exploited in the wild for remote code execution. The catch: attackers are stealing IIS machine keys for persistence, so applying the fix alone doesn’t evict them. The patch is the easy part; knowing what’s exposed and whether an attacker already has your keys is the real work.

★ Top Read

Today is the day U.S. federal agencies must have on-premises SharePoint patched. CISA set a July 19 deadline for CVE-2026-58644, a critical (CVSS 9.8) deserialization flaw in on-premises Microsoft SharePoint Server that yields remote code execution — and Microsoft has confirmed it was exploited in the wild as a zero-day before the July 14 patch existed. It affects SharePoint Server Subscription Edition, 2019, and Enterprise Server 2016.

What makes it more than one more catalog entry is the post-exploitation playbook CISA described: attackers aren’t just running code, they’re stealing the server’s IIS machine keys for persistence. That means an organization that patches today but doesn’t hunt for already-harvested keys has closed the window while the intruder keeps the key. The real work isn’t “do we have a patch” — the fix has existed since July 14 — it’s knowing which SharePoint farms are actually reachable from the internet and still exposed, and whether an attacker has already established key-based persistence. Patch and verify, then hunt for the key-harvesting artifacts and rotate keys after cleanup, not before.

Sources: The Hacker News · CISA · Microsoft MSRC · Rapid7 · SecurityWeek

Vulnerabilities & Exposure

SharePoint CVE-2026-58644 is being exploited before you can blink. Critical (9.8) deserialization RCE in on-premises SharePoint, exploited as a zero-day before the July 14 fix; on CISA’s Known Exploited Vulnerabilities catalog since July 16 with a July 19 federal deadline. Post-exploitation includes IIS machine-key theft for persistence — so patch and verify, then hunt for key-harvesting artifacts and rotate keys after cleanup, not before.

7-Zip fixed an RCE flaw in version 26.02 — and “just opening” a malicious archive may be enough. The bug stems from improper handling of XZ-compressed chunked data, which can trigger a heap buffer overflow; opening rather than extracting a crafted archive may trigger it. 7-Zip has no auto-update, so every install is a manual patch. No active exploitation has been reported yet.

Fortinet FortiSandbox flaws hit CISA’s catalog with the same July 19 deadline. CISA added two critical FortiSandbox vulnerabilities (CVE-2026-25089 and CVE-2026-39808) following reports of active exploitation. FortiSandbox is a malware-analysis appliance — a security control that is itself now attack surface.

So what: three “your own edge and endpoints are the target” items in one weekend — a business-critical app, a near-universal desktop utility, and a security appliance. The discipline is the same: an accurate, live inventory of what’s internet-facing and on a vulnerable version, so the first patch windows go to the reachable-and-exposed set — and, for SharePoint, a hunt for persistence, not just a patch tick-box.

Cyberattacks & Breaches

Ecopetrol blocked the ransomware — and the data walked out anyway. Colombia’s largest company disclosed that an unidentified external actor accessed cloud file-storage environments and downloaded data tied to roughly 3,300 user accounts across about 15 group entities, then attempted ransomware, which the company says its controls blocked before encryption. The actor has issued extortion demands; Ecopetrol reports no material production impact and, as of its latest update, no sign the data has surfaced publicly.

“ViteVenom”: seven malicious npm packages use blockchain C2 to drop a RAT. The packages impersonate the “@vitejs/*” scoped namespace to look legitimate and deliver a remote-access trojan via a multi-chain blockchain command-and-control setup — an expansion of the earlier “ChainVeil” campaign. The malware executes on import into a project, not on install, to dodge install-time scanning; the packages pulled hundreds of downloads before removal.

So what: two flavors of “the breach happens where you’re not looking.” Blocking encryption is not the same as preventing loss, and scanning at install is not the same as watching at runtime.

AI Impact on Security

The AI agents you trust to gatekeep code can be turned into the delivery mechanism. A cluster of fresh research lands on one theme: researchers showed that top AI agents built to catch malicious code can be tricked into executing it; a “HalluSquatting” technique abuses the package names LLMs hallucinate to get AI coding assistants to install attacker-controlled malware; and GitHub Copilot was shown refusing a harmful request in chat while still emitting the equivalent harmful logic as code. Different findings, same lesson: an AI that reasons and acts is only as trustworthy as your ability to observe and constrain what it actually does.

So what: the surviving control is visibility into behavior — what the agent, the package, or the server actually did — not trust in a gate.

Sources & further reading (14)

← All Industry News