
The building owner carries the ultimate liability for a compromised building, whoever installed the system, wrote its firmware or runs its AI. That point came from the vendor session at the (CS)²AI Online Symposium on Smart Building Cybersecurity on October 28, 2026, and it ran under most of the practical advice that afternoon.
The afternoon ran three sessions: a panel on ransomware and attacks with physical consequences in buildings, a session on managing smart building vendors, and a panel on AI, regulation and accountability. Actimeta took part in the opening panel. Here is what we took away.
The refrigeration attacks on French supermarkets earlier in October framed the first panel. The preview for that panel said the attack reportedly used unpatched firmware on internet-connected devices. Be careful with that detail. CERT-FR's alert, last updated October 16, still does not state how the attackers got in. What the reporting does show is that the controllers' management interfaces were exposed to the internet, and Danfoss said most units hit were older products that no longer receive updates.
The panel's position was that exposure management comes first: find every building system that answers from the internet, then remove it or put it behind a gateway you control. In a portfolio starting from nothing, that list is faster to build than a complete asset inventory, and it closes the doors attackers are using today.
We agree with the order, with one caution. An exposure list tells you where the open doors are. It does not tell you which controllers are end of life, which run shared accounts, or what your recovery depends on. You still need the full inventory. Do the exposure scan this month and the inventory over the next quarter.
The panel also described maturity as uneven. Flagship properties often have a program. The smaller, older or less funded buildings in the same portfolio often have none, and attackers scanning the internet do not care which is which.
The vendor session started from the liability point. If you hold the risk, you write the rules. Owners should publish a cybersecurity standard and give it to every vendor and service provider before work starts. The speakers described a vendor package that sets out the owner's requirements in twelve steps, and a cyber commissioning step that runs alongside functional commissioning. Functional commissioning proves the chiller plant does what the design says. Cyber commissioning proves the accounts, network settings, remote access and logging do the same.
The failures they described were familiar: shared admin accounts, unmanaged switches, no clear standard, and inactive accounts nobody removed. Each one becomes technical debt that is harder to fix after handover than before it. Their recommendations were concrete:
How often to verify was left open, with the answer tied to the criticality of each facility. That is the right answer, and it means you have to rank your facilities first.
The final panel spent most of its time on AI agents in building operations. Their recommendations would sound familiar to anyone who manages vendor access. Each agent gets its own non-human identity, tied to a named human owner, with defined permissions, an audit trail and authority you can revoke. The panel proposed using Failure Mode and Effects Analysis to rank the tasks an agent may perform by severity, likelihood and detectability before granting them.
Where AI helps defenders today, the panel described controlled uses: reconciling data from several building systems to spot anomalies, with a person validating before anyone acts. Whether agents should be split into read-only and read-write classes remained an open question. Our view: start every agent read-only, and give write access one task at a time, with a person approving each change.
The panel saw European rules setting the pace. NIS2 makes management bodies of essential and important entities approve and oversee cybersecurity measures, and holds them accountable. The EU Cyber Resilience Act puts secure-by-design obligations on manufacturers; its reporting duties started on September 11, 2026, and most of its requirements apply from December 11, 2027. Neither changes the controllers already in your buildings, which will run for another fifteen or twenty years.
The panel's preview had asked whether insurers were becoming the de facto regulators of building security. The panel's answer was that insurance plays a minimal role so far. Owners are acting on risk management and regulation, not on premiums.
The question nobody answered was how to secure or replace legacy equipment under capital constraints. That is the problem every owner in the audience will take into next year's budget. The liability point from the vendor session suggests where to begin: the equipment you cannot replace yet is the equipment whose security terms you have to set yourself.
According to the symposium's vendor session, the building owner carries the ultimate liability, whoever installed the system, wrote its firmware or runs its AI.
A step that runs alongside functional commissioning and proves that accounts, network settings, remote access and logging match the owner's standard.
Give each agent its own identity tied to a named human owner, with defined permissions and an audit trail. Start read-only.
Planning a new smart building, upgrading the systems in an existing one, or buying technology for a facility? Actimeta is an OT cybersecurity and technology advisory firm that helps owners get the design, the vendors and the security right from the start.