Thursday, October 1, 2026|Issue 14|6 min read|Join the discussion
Security is increasingly the management of delegated authority
Email infrastructure is back in the incident-response queue, AI agents are moving from software abstraction into purpose-built infrastructure, and regulators are starting to ask who is responsible when autonomous systems cross boundaries. The common thread is delegated authority.
The short version
- Microsoft details active exploitation of Zimbra CVE-2026-73570: a crafted email can trigger command injection on servers with zimbra-snmp and SNMP notifications enabled.
- NetScaler's federal deadline has passed; exploitation is confirmed, so investigation is the work now.
- CoreWeave says Cognition is the first customer running production workloads on NVIDIA Vera Rubin NVL72.
- Google's Gemini 4 Argon is rolling out first to a set of trusted cyber defenders.
- The FTC has opened a broad probe into AI safety at Anthropic and OpenAI, according to news reports.
Issue at a glance
2 critical
Breaking
Zimbra / email / KEV
A crafted email can own a Zimbra server
Microsoft Threat Intelligence tracked exploitation of CVE-2026-73570, an unauthenticated OS command injection in the Zimbra Collaboration Suite SNMP notification path. On internet-facing servers with the optional zimbra-snmp package installed and SNMP notifications enabled, a specially crafted email can trigger it without authentication or user interaction. Observed activity included JSP web shells, reverse shells, privilege escalation, persistent remote tooling, memory-backed execution and mailbox data collection. It affects ZCS before 10.1.20 and has been on CISA's KEV list since August.
Why it matters. The mail server sees everything and is reachable by anyone who can send it a message.
What to do. Identify servers with zimbra-snmp, confirm 10.1.20 or later, restrict exposure and hunt for historical compromise: web shells, unexpected child processes of mail services and archive-and-transfer activity.
CVE-2026-73570
Citrix / NetScaler / incident response
NetScaler: the deadline passed, the investigation didn't
The September 30 federal deadline for CVE-2026-88771 and CVE-2026-88772 is behind us. Citrix and CISA confirm exploitation in the wild, and Unit 42 recommends isolating vulnerable systems and preserving snapshots and logs.
Why it matters. Patching ends the vulnerability window, not the intrusion.
What to do. Preserve logs and artifacts, run integrity and IOC checks, and investigate authentication events, configuration changes and outbound connections from the appliance.
CVE-2026-88771, CVE-2026-88772
Citrix security bulletin CTX697096|CISA|Unit 42|LevelBlue SpiderLabs
Trends
What can this thing cause another trusted system to do?
Security is increasingly the management of delegated authority: the mail server, the service principal, the CI/CD pipeline, the RMM tool, the AI agent, the edge appliance. Each one can make another trusted system act. Ask of each: what can this thing cause another trusted system to do?
Agent security is becoming infrastructure security
CoreWeave says Cognition, maker of Devin, is the first customer anywhere running production workloads on NVIDIA Vera Rubin NVL72, citing agentic workloads with long contexts and thousands of concurrent tasks. Google announced Gemini 4 Argon, rolling out first to trusted cyber defenders, and says it is taking part in the US government's voluntary pre-release access process. Agents are getting dedicated hardware and real authority; their identities and permissions need the same rigor as any privileged service.
Regulators start asking who is responsible
The Washington Post and Bloomberg Law report that the FTC has opened a broad investigation into the safety of AI systems from Anthropic and OpenAI, citing unnamed officials. The probe has not been made public and its scope is unclear.
Behavior detections outlast CVE detections
A mail server spawning a shell is suspicious whatever the CVE. Zimbra's chain is a reminder that the vulnerability database tells you less than the attack chain, and that behavioral SOC detections keep working after the next bug lands.
Clearance corner
Risk-driven remediation under BOD 26-04
Zimbra and NetScaler both show why BOD 26-04 prioritizes known exploitation and exposure. Being able to explain the attack path and how remediation was verified matters more than reciting a CVSS score.
What we are watching next
Zimbra
More victims and IOCs for CVE-2026-73570.
NetScaler
Post-deadline compromise findings.
Vera Rubin
How fast purpose-built agent infrastructure scales beyond first customers.
AI-agent governance
Scope and formal demands in the FTC probe.
Agent Zero Trust
Identity, least privilege and audit for autonomous agents.
Sources
Join the discussion
Loading comments…