Today’s digest is a reminder that “patched” doesn’t always mean “you had a choice in the matter.” One story below is a cloud-hosted identity flaw Microsoft fixed on its own infrastructure with zero customer action possible, another is a malware campaign still exploiting a bug that’s had a patch since June, and the third is an AI assistant flaw that needed no code execution at all — just a link and a click. Here’s what’s new since yesterday’s digest.
Identity watch: A maximum-severity remote code execution flaw in Microsoft Entra ID was exploited before Microsoft’s silent fix
Microsoft disclosed on August 20 that CVE-2026-69836, a CVSS 10.0 unsafe-deserialization flaw in Entra ID — the identity platform behind sign-in for Microsoft 365, Azure, and most connected third-party apps — was exploited in the wild before it was patched. The bug let an attacker send specially crafted serialized data to a vulnerable backend endpoint and get arbitrary code execution with no authentication and no user interaction required. Because Entra ID is a fully managed cloud service, Microsoft rolled the fix out on its own infrastructure; there’s no update package, KB article, or configuration change for customers to apply, and Microsoft says the flaw is now fully mitigated. How attackers found and used it before disclosure hasn’t been detailed. The Hacker News
A CVSS 10.0 flaw in the service that authenticates almost everything is about as high-stakes as identity infrastructure risk gets, and the fact that there’s nothing to patch doesn’t mean there’s nothing to do. If your organization relies on Entra ID, this is a good prompt to review sign-in and token-issuance logs from the affected window for anomalies — unexpected token grants, impossible-travel sign-ins, or privilege changes you can’t otherwise explain — since a silently-patched backend flaw gives you no other way to confirm you weren’t touched.
Actively exploited watch: CISA adds two TrueConf Server flaws tied to an ongoing trojanized-installer campaign
CISA added CVE-2026-72529 (missing authentication for a critical function) and CVE-2026-72530 (code injection) in TrueConf Server to its Known Exploited Vulnerabilities catalog on August 20, giving federal agencies until early September to patch or discontinue use. The pair lets an unauthenticated attacker with network access to TCP port 4307 run an arbitrary script and then break out of the resulting isolated environment to execute code on the host. The vulnerable chain has been under active abuse since at least this spring by a group researchers track as PhantomCore (linked to the Head Mare activity cluster), which has been compromising unpatched TrueConf video-conferencing servers and replacing legitimate client installers with trojanized versions carrying a backdoor and remote-access trojan. TrueConf shipped fixed versions 5.3.9, 5.4.9, and 5.5.5 back in June. Kaspersky ICS-CERT · BleepingComputer
This is a two-month-old patch still being actively exploited, which makes it a supply-chain risk as much as a server-hardening one — anyone who pulled a client installer from a compromised TrueConf deployment during that window could be running a backdoor regardless of whether their own server is patched. If you run TrueConf anywhere in your environment, confirm you’re on 5.3.9/5.4.9/5.5.5 or later, and if any client installers were distributed from an unpatched server since spring, treat those endpoints as suspect until verified clean.
AI security watch: A chained Copilot Personal flaw let one click exfiltrate data from connected accounts
Researchers at Varonis disclosed CVE-2026-24301 (CVSS 8.8), a set of three chained weaknesses in Microsoft Copilot Personal they call CoSnitch. Chained together, the flaws let an attacker get a victim to click a single crafted link, which silently triggered Copilot to run an attacker-controlled prompt inside the victim’s own authenticated session — no further interaction needed. From there, the technique could pull data out of connected Gmail or Outlook accounts, Google Calendar, Google Drive, and Copilot’s own chat history and persistent memory, and could also be used to poison that memory with false information. Researchers found the underlying auto-execute mechanism, an undocumented ?autorun=1 URL parameter, by repeatedly questioning Copilot about its own constraints rather than analyzing code directly. Varonis reported the issue in December 2025; Microsoft fixed it on August 18 and says enterprise Copilot customers were never affected, only the consumer Copilot Personal product. No evidence of exploitation before the patch has surfaced. The Hacker News
The interesting part isn’t just the bug — it’s the discovery method, since “ask the model how it works” is now a viable recon technique against any AI assistant with enough conversational latitude, not just Copilot. If your team evaluates or deploys AI assistants with access to connected accounts (email, calendar, storage), this is a fair prompt to check whether similar auto-execute or session-continuation parameters exist in your own stack, and whether a single malicious link could act with a user’s full authenticated context.
Final thought
Three different flavors of “patched” today: an identity platform fixed invisibly by the vendor with attackers already inside the window, a two-month-old server patch still losing ground to an active trojanized-installer campaign, and a consumer AI flaw closed before anyone but the researchers who found it seems to have used it. The common thread is that patch status alone doesn’t tell you whether you were exposed — for the Entra ID and TrueConf stories in particular, the only way to know is to go back through your own logs. If you want help sizing up what your logging actually shows for a window like that, see our how-to guides or book a discovery call.