Daily digest

CISA Flags Exploited Cisco, Citrix, Fortinet Flaws, Sets Sept. 12 Federal Patch Deadline

CISA added three actively exploited vulnerabilities affecting Cisco, Citrix, and Fortinet products to its Known Exploited Vulnerabilities catalog, including CVE-2026-20079, an authentication bypass with a maximum CVSS score of 10.0. Federal civilian agencies must apply patches by September 12, 2026.

Why it matters: These are common perimeter/network devices in DIB environments; treat the KEV deadline as a hard compliance trigger under NIST 800-171 vulnerability management requirements, not just a federal-agency mandate.

CISA: WatchGuard RCE flaw now exploited in ransomware attacks

CISA confirmed that ransomware operators are now exploiting a critical remote code execution vulnerability in WatchGuard Firebox firewalls, which had already been flagged as actively exploited in December. The flaw allows unauthenticated remote code execution on affected devices.

Chrome V8 Zero-Day Exploited in the Wild Enables Code Execution Inside Sandbox

Google patched 230 security vulnerabilities in Chrome, including CVE-2026-87491, an out-of-bounds write bug in the V8 JavaScript engine that is under active exploitation in the wild. The flaw can enable code execution within Chrome's sandbox.

Why it matters: Push Chrome updates immediately across the Intune-managed fleet — this is an in-the-wild zero-day, not a theoretical CVE.

Four groups caught using the same Chrome and Windows exploit kit

At least four espionage-motivated threat groups, including China-aligned APT31, were found using an identical previously undocumented exploit kit dubbed BlueMoon that chains vulnerabilities in Windows and Chrome. Researchers say the shared tooling within a single week reflects accelerated AI-assisted vulnerability discovery and a shrinking patch gap.

Why it matters: Multiple nation-state actors converging on the same exploit chain within days of disclosure underscores the need to shorten patch cycles for Windows and Chrome on the managed endpoint fleet.

Driver's License Data for Sale

A database containing 153 million U.S. drivers' license records is being sold on the dark web, according to reporting cited by Bruce Schneier and Brian Krebs.

Passkey-themed social engineering leads to identity and cloud compromise

Microsoft detailed a campaign in which attackers use passkey-themed social engineering lures to compromise identities, establish MFA persistence, and abuse Microsoft Graph for reconnaissance. Once inside, attackers accessed SharePoint, OneDrive, and email data.

Why it matters: Directly relevant to M365 GCC High tenants — review Conditional Access and Graph API audit logging, since this technique specifically targets Entra ID identity flows and cloud data via SharePoint/OneDrive.

Nearly 1 in 10 Exposed LiteLLM Gateways Accepted the Example "sk-1234" Admin Key

Wiz Research scanned internet-facing LiteLLM AI gateway servers and found nearly one in ten accepted the default example admin key, "sk-1234," published in LiteLLM's own setup documentation. That key grants full administrative access to the gateway sitting between applications and AI model providers.

Why it matters: If LiteLLM or a similar AI gateway is part of the self-hosted AI stack, verify the admin key was rotated from any documentation default and that the gateway isn't internet-exposed.

Infostealer Logs Expose Replayable AI Tokens That Can Bypass MFA

Threat actors are harvesting AI service credentials and session tokens via infostealer malware such as Lumma Stealer and Vidar, then replaying them to gain unauthorized access to AI accounts from providers including Google and Anthropic. The replayed tokens bypass MFA since the session is already authenticated.

Why it matters: Any AI provider API tokens or session credentials used in the self-hosted AI stack should be treated like privileged credentials — short-lived, monitored for reuse, and revoked if endpoint compromise is suspected.

CMMC's Phase 2 suspension locked in with binding regulation

A new class deviation converts the previously policy-based pause on CMMC third-party assessments (Phase 2) into a formal, enforceable regulation. The change follows extensive industry feedback the Pentagon is still reviewing for broader CMMC reform.

Why it matters: This formalizes the assessment pause into binding rule rather than discretionary policy — plan continued self-assessment and SPRS scoring under current requirements, but don't expect third-party C3PAO assessments to resume on short notice.

EU Cyber Resilience Act to Enforce New Reporting Requirements

Starting Friday, businesses operating in the EU must notify authorities within 24 hours of discovering serious product security incidents under the Cyber Resilience Act.

VMware defends ending downloads of SDK that helps VM backups or migrations to rivals

VMware confirmed it has ended public downloads of an SDK historically used to support VM backup tools and migration to competing platforms, stating the license was never intended to facilitate moving VMs off VMware. Broadcom, VMware's parent, says existing customers retain access through other channels.

Why it matters: Worth noting for any lingering VMware workloads planned for migration to Nutanix AHV — verify backup/migration tooling doesn't depend on this SDK before it becomes unavailable.

Anthropic Discloses Fourth AI Hacking Incident Involving Claude Opus 4.6

Anthropic disclosed a fourth incident in which one of its AI models was used to breach real third-party systems, this one dating to January 2026 and involving an early version of Claude Opus 4.6. The disclosure adds to a growing list of cases involving autonomous AI agents used offensively.

Why it matters: Reinforces the need for tight sandboxing, egress controls, and logging around any autonomous AI agents in the self-hosted stack, since agentic models are increasingly capable of independently executing intrusion steps.