Zentera News

Threat Briefing – November 2026

Written by Zentera Press | Nov 17, 2026, 6:53:29 PM


In this month’s Threat Briefing, we look at three recent incidents: an AI agent that broke into a security nonprofit in seconds, a sabotage campaign against supermarket refrigeration across Europe, and a firewall that became the way in for a wiper attack. The attackers and their methods have little in common. What they share is the part after the first system fell. In each case, how much damage followed came down to whether the next system was reachable.

An AI Agent Breaks Into a Vulnerability Disclosure Nonprofit

On September 21, attackers broke into the Dutch Institute for Vulnerability Disclosure (DIVD), a volunteer nonprofit that finds exposed systems and warns their owners. DIVD assessed that an autonomous AI agent ran the attack. The agent chained two previously unknown flaws in Zammad, the open-source help desk platform DIVD uses: a session hijack that gave it code execution, then a privilege escalation to root. DIVD says the trip from hijacked session to root took seconds. The agent was noisy and careless. It left comments explaining its own logic in its code, and a password spray it ran broke its own adversary-in-the-middle setup. It still took data, including volunteers’ email addresses. What it could not do was go much deeper. DIVD credits network segmentation and its incident response team with limiting how far the attacker got.

Nobody can patch a flaw that nobody knows about, and no human responder reacts as fast as an agent. Segmentation depends on neither. When each application runs in its own virtual chamber with identity-based policy, root on the help desk host gives an agent no route to the next system, because that route exists only if policy creates it. An agent can test every adjacent path at machine speed and still find nothing to connect to. [See how Virtual Chambers stop lateral movement →]

Freezers Set to 50°C Across Europe

Overnight on October 11 to 12, attackers took control of refrigeration controllers at about 700 French supermarkets and forced them into defrost, pushing cabinet temperatures as high as 50°C. Chains including E.Leclerc, Intermarché, Carrefour, Monoprix and Coopérative U threw out chilled and frozen stock, French authorities counted at least 15 fires, and losses run to tens of millions of euros. On November 2, supermarkets in several more European countries, including Belgium, Sweden, Estonia, Ireland and Luxembourg, reported the same kind of tampering. The same night, a refrigeration supplier to France’s national blood service (EFS) was compromised, and temperature and defrost thresholds in cold rooms at three plasma storage sites were raised to extreme levels. EFS says its own monitoring and alarms caught the change before any product was lost.

The attackers didn’t need anything exotic. The controllers named in France’s government alert, Danfoss AK-SM and AK-SC units and Copeland and Dixell XWEB units, were reachable from the internet, many of them on outdated or end-of-life firmware. Danfoss said the Dutch systems hit in November weren’t on a secure network and still used default passwords. On November 9, France formally attributed the October wave to Shahid Kaveh, a cyber unit of Iran’s IRGC. Russian hacktivist groups have claimed attacks across Europe, and the November wave has not been formally attributed.

These controllers should never have been reachable from the internet. When OT assets sit inside a chamber, there is nothing exposed to scan and nothing listening. Contractors and administrators connect through a policy-controlled chamber with MFA and time-limited access to the one controller they need. That matters here, because supermarkets and the blood service alike relied on outside refrigeration suppliers to manage these systems. [See how Zentera secures OT environments →]

A Firewall Opens the Door to an ESXi Wiper

On November 12, Check Point disclosed attacks on unpatched Security Gateways. The attacker got in through PHP2Shell flaws in the Mobile Access portal, then used a previously unreported privilege escalation bug in Gaia OS (CVE-2026-123039) to get root. From there it planted web shells and a backdoor called TSGA and scanned the internal network. In some intrusions it moved over SSH to VMware ESXi hosts using the vpxuser service account and ran a wiper called destroy_v6. Check Point saw two waves: one starting October 30 aimed at Israeli organizations, which included the destructive activity, and a broader one on November 2 with no destruction observed. CISA added the privilege escalation flaw to its Known Exploited Vulnerabilities catalog on November 13. The attacker has not been publicly identified.

The pattern is familiar. A perimeter device falls, and the attacker works inward until it reaches the hypervisors or other critical systems. If ESXi hosts and management interfaces live in their own chamber, a compromised gateway has no path to them. SSH from a gateway address to an ESXi host is not permitted by policy, so it never connects. Remote users reach specific applications, not the whole network. [See how Universal ZTNA limits access to applications →]

The Pattern

An AI agent, an OT sabotage campaign, and an edge device exploit. Three different attackers, and in all three, getting in was cheap. How far they got afterward depended on what was reachable from where they landed. DIVD had segmentation, and the damage stayed limited. The refrigeration controllers sat on the open internet. The compromised gateways could reach the hypervisors.

Attackers, human or AI, now move at machine speed, so defense can’t rest on patching faster or detecting sooner. The fix is the same across all three:

  • Chamber your applications so a compromised host has no route to its neighbors.
  • Take OT off the internet and put contractor and admin access behind a policy-controlled chamber.
  • Assume the edge device will fail, and keep hypervisors and management interfaces out of its reach.
  • Give your own AI agents least-privilege chambers so they reach only what they’re meant to.

Key Takeaways

  • Machine speed needs a path. An AI agent with root on one host still can’t move if policy never created a route to the next system.
  • Exposure is the starting point. Internet-facing OT controllers with weak or default credentials gave attackers everything they needed. Take the exposure away and the attack has nowhere to begin.
  • The edge will fail. Treat every perimeter device as eventually compromised and make sure it has no path to your hypervisors or management planes.

If you want to see how chambers would apply to your applications, OT, or edge devices, reach out and we’ll connect you with one of our solutions engineers.