Field Notes — Week of August 2, 2026
A short, opinionated digest of what actually mattered this week across cloud, security, AI, and automation — and what it means if you run real systems.
Security: “Certighost” turns an AD CS enrollment quirk into domain takeover
Researchers H0j3n and Aniq Fakhrul dropped a working proof-of-concept for CVE-2026-54121 (“Certighost”) — an Active Directory Certificate Services elevation-of-privilege flaw rated CVSS 8.8 that Microsoft patched on July 14, with the PoC public by July 24. The exploit lets any authenticated domain user — no admin rights — impersonate a Domain Controller and walk out with a forged DC certificate.
The mechanism is an enrollment fallback (a “chase”) during directory-object resolution: by supplying request attributes like cdc, an attacker coaxes the CA into pulling identity data from an attacker-controlled host and issuing a certificate on that basis. From there it’s krbtgt extraction and full domain compromise. It hits Enterprise CA deployments across Windows Server 2012 through 2025 — essentially any org running internal Microsoft PKI.
The take: A low-privileged-user-to-domain-admin path with a public PoC is a drop-everything item. If you run AD CS, confirm the July cumulative update is actually applied to every CA — not just scheduled — this week. Then audit for the misconfiguration that makes the chase exploitable, and hunt certificate-issuance logs for DC-template certs requested by non-DC principals since mid-July. Your PKI is a domain-takeover primitive; treat CA servers like Tier 0.
Automation: an unauthenticated RCE lands in the CI/CD server itself
JetBrains shipped an emergency fix for CVE-2026-63077, a CVSS 9.8 unauthenticated remote code execution flaw in TeamCity On-Premises. The bug is a deserialization weakness in the agent polling protocol: an attacker with only HTTP(S) access can bypass authentication and run arbitrary OS commands as the TeamCity server process. All On-Premises versions are affected; the fixes are 2025.11.7 and 2026.1.3, with a patch plugin for anyone who can’t upgrade immediately. Cloud customers need no action.
Reported privately on July 10 by Antoni Tremblay, it was not known to be exploited at disclosure — but a build server is a high-value target, and this one leaks stored credentials and lets an attacker rewrite pipeline output.
The take: Your CI/CD server holds deploy keys, cloud credentials, and signing material — compromise it and every downstream system is in play. Patch TeamCity now, then treat the build server as the Tier 0 asset it actually is: no direct internet exposure, egress-filtered, credentials rotated on the assumption they may already be gone. If it was reachable before you patched, rotate the secrets it stores before you call it done.
Cloud: Google Cloud makes Looker MFA the default
Quieter week for keynotes, but a meaningful default flipped: Looker (original) instances on 26.12 can now require MFA for email-and-password logins — and it’s enabled by default (rolled out July 27–30). Google also GA’d Looker’s custom calendar, Expression Assistant, and enhanced search, and shipped a new Managed Service for Apache Airflow release on July 29.
The take: Secure-by-default is the right direction, but “enabled by default” is exactly the setting that surprises admins mid-quarter. If you run Looker with local logins, get ahead of it: confirm the MFA rollout, communicate to users before they hit an enrollment wall, and — better — use this as the nudge to move Looker auth behind your SSO/IdP so password logins stop being a thing you have to secure at all.
AI: Cognizant targets the 88% agent-pilot failure rate
Cognizant launched an EMEA AI unit on July 28 built specifically to drag agentic AI from stalled pilots into production, citing the IDC-documented 88% pilot failure rate. The model is deliberately multi-cloud and multi-vendor — a “Frontier Deployed Engineering” approach that helps clients pick and scale the stack that fits rather than committing to one vendor.
The take: The failure rate isn’t a model-quality problem — it’s an integration, governance, and change-management problem. The pilots that die never had eval harnesses, human-in-the-loop escalation, or scoped credentials. Before you greenlight another agent POC, define the production bar first: how it’s monitored, what it’s allowed to touch, and who’s in the escalation path. Build backward from that, and design for portability so you’re not locked into one vendor’s agent glue.
ZelMor LLC helps organizations and consultancies with cloud architecture, AI enablement, security & IAM, and automation. If any of the above is a live question for your team, start a conversation.