Certighost: A Public Exploit Just Turned Any User Into a Domain Controller

Certighost AD CS domain-takeover exploit goes public (CVE-2026-54121). A low-privileged AD user can drive the CA to issue a certificate impersonating a Domain Controller, then authenticate as that DC and DCSync the domain (including the krbtgt hash). Fixed in Microsoft’s July 2026 security updates; the fix is a servicing-flag-gated validation that the enrollment “chase” target is a genuine DC computer object. With a functional exploit now circulating, the window between “patched Patch Tuesday” and “weaponized in the wild” is closing this weekend. (Microsoft July 2026 security updates / MSRC; The Hacker News; Cyber Security News)

So what: this is an identity-fabric emergency, not a routine CVE. Confirm the July 2026 cumulative update is deployed across every Certification Authority and domain controller, then verify the mitigation is actually enabled (a patch present but the fix dormant behind a feature gate is the trap here); after that, treat detection as the real control — alert on certificates issued with DC/machine identities, on unusual enrollment attributes, and on DCSync/replication from anything that is not a DC.

Check Point SmartConsole auth-bypass (CVE-2026-16232) — CISA fix-by date is today. A critical (CVSS 9.3) improper-authentication flaw in Check Point Security Management / Multi-Domain Management lets an unauthenticated remote attacker obtain an application login token and act with full administrator privileges, changing security policy at will. Check Point shipped a Jumbo Hotfix on July 22 and confirmed exploitation; CISA added it to the Known Exploited Vulnerabilities catalog on July 22 with a Federal Civilian Executive Branch remediation deadline of July 25 — today. (Check Point advisory; CISA KEV; Rapid7; BleepingComputer)

So what: an admin-plane takeover on the console that governs your firewall policy is as bad as it sounds — the attacker doesn’t need to break the firewall if they can rewrite its rules. Any org running Security Management or MDS should treat the hotfix as same-day, confirm the token-issuance path is patched, and hunt for unexpected policy changes or admin sessions since mid-July.

Russian “LAUNDRY BEAR” espionage crew reads NATO and government mail via a zero-click Zimbra flaw. A joint cybersecurity advisory (AA26-204A), issued July 23 by CISA and partners, warns that a Russian state-supported group — tracked as LAUNDRY BEAR / Void Blizzard — has been exploiting a Zimbra Collaboration webmail vulnerability (CVE-2025-66376, an XSS from insufficient sanitization of CSS @import directives) to steal email with no link-click required: simply viewing the malicious message triggers exfiltration. The payload pulls the last ~90 days of mail, the organization’s entire address directory, browser-saved passwords, and two-factor recovery codes, and has hit government, defense, energy, technology, education, media, and NGO targets since at least mid-2025. The flaw was patched in November 2025, but the actors operationalized it earlier, making it a true zero-day at first use. (CISA advisory AA26-204A; joint CSA via defense.gov; The Hacker News)

So what: a zero-click mail-read that also lifts 2FA recovery codes is an account-takeover engine, not just a data-theft one. If you run Zimbra, confirm you are on a build patched after November 2025, assume anything unpatched between mid-2025 and now was readable, and rotate credentials and 2FA recovery codes for exposed mailboxes rather than only patching. The durable lesson: “no click required” is now the baseline threat model for high-value inboxes.

UAC-0099 hides malware inside a fake Notepad++ plugin to hit Ukrainian orgs. CERT-UA warned that the Russia-aligned UAC-0099 cluster has swapped tactics: a phishing lure leads to a ZIP bundling a full legitimate copy of Notepad++ 8.8.3 plus a trojanized plugin, where a rogue NppExport.dll sideloads an upgraded MATCHBOIL.V2 loader (alongside newer LUNCHPOKE and BURNYBEAR tooling). The loader establishes scheduled-task persistence, updates C2 config, and pulls further payloads — abusing a trusted developer utility as the cover for execution. UAC-0099 has previously served as an initial-access provider for Sandworm/APT44-linked operations. (CERT-UA via The Hacker News; Security Affairs; BleepingComputer)

So what: DLL-sideloading through a legitimately-signed, widely-installed dev tool is designed to slide past both users and allow-listing. Detection lives below the filename: watch for Notepad++ loading plugins from unexpected paths, scheduled tasks created shortly after archive extraction, and outbound C2 from a process users assume is benign. It’s a Ukraine-targeted campaign today, but the plugin-sideload pattern travels.

The week’s identity-abuse pattern is the one agentic AI will stress next: who is allowed to be whom. Read the three technical stories above together and one control keeps recurring: Certighost forges a machine identity, Check Point forges an admin token, and the Zimbra crew steals the recovery codes that would let you re-establish identity after a reset. The through-line is authorization and identity, not any single CVE — and it is exactly the surface that multi-agent AI deployments expand. An AI agent handed a credential and a task is one more principal that can be impersonated, over-permissioned, or coaxed outside its lane. It isn’t the model (or the CA, or the console) that matters, it’s the harness around it: can you constrain what each identity is allowed to do, and can you see when one steps outside it? (OWASP LLM/agentic-AI security guidance; Help Net Security)

So what: the least-privilege-and-behavioral-telemetry boundary that would have caught a forged DC certificate is the same boundary that governs an autonomous agent touching real infrastructure. If your identity program still treats “machine,” “admin,” and “agent” as separate afterthoughts, this week is the argument for unifying them under one visibility-and-control plane — before you add a fleet of AI principals to the mix.

EU AI Act transparency obligations enforce August 2; Commission published implementing guidelines July 20. Providers and deployers of chatbots, synthetic-media generators, and emotion-recognition systems must disclose AI interaction and make AI-generated content identifiable. It pairs with the July 7 EU Action Plan on Cybersecurity and AI, under which ENISA and the Commission’s Joint Research Centre are building a secure platform to test AI models against simulated cyber-threats and to evaluate advanced models before they reach the EU market. (European Commission — AI Act; EU Cybersecurity-and-AI Action Plan)

So what: the near-term action for US enterprises is disclosure hygiene on any customer-facing AI feature touching the EU market before August 2 — eight days out. The durable one is evidentiary: being able to show what your AI systems did and labeled is becoming a compliance obligation, the same “prove what happened” capability this week’s incidents keep demanding.