Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

Cyber risk in transport is usually framed around a collision. Signals manipulated, a train sent onto occupied track, an accident caused deliberately. ‍ Signalling is built to make that outcome unavailable. Any failure forces the system into its most restrictive state, so a compromise produces a halt rather than a crash. The loss is the halt, and rail is unusual in already having a published price for one. ‍

A cyber loss model has dozens of inputs and every one of them can be argued about. Record counts, downtime costs, control effectiveness, secondary loss factors, event likelihoods. ‍ A few of them determine the answer and the rest barely move it. Knowing which is which tells you where estimation effort belongs, and more usefully which disagreements about the model are not worth having. ‍

A model produces a figure, an event happens, and somebody asks whether the figure was right. It is the obvious question and it has almost no published answer, because the comparison is harder than it looks. ‍ A single realized loss cannot falsify a distribution. If a model puts a one percent chance on exceeding a threshold and the threshold is exceeded, the one percent case occurred, which is what the model said would sometimes happen. ‍

The word agent appears nowhere in the AI Act. Some read that absence as a scope question still to be settled, and treat agentic deployments as sitting outside a regime written before they existed. ‍ On its own FAQ the Commission has answered it directly. Agents are not a separate category and the existing definitions already reach them, so nothing needs amending for the rules to apply.

An interruption model measures the time from failure to restoration. Systems down, systems back, multiply by revenue per hour. ‍ In an airline the outage ends well before the recovery does, and the ratio between the two is large enough to make a model keyed to restoration wrong rather than imprecise. One carrier restored connectivity in under an hour and the resulting displacement ran into the following morning. ‍

A correlation rule joins several telemetry sources and fires when they agree. Guidance on writing them concentrates on thresholds, ordering and tuning for noise. ‍ The decision that determines whether a rule works is upstream of all of that. Each source in a rule plays one of three roles, and treating them interchangeably is what produces a rule that misses real events or fires on ones nobody can act on. ‍

A breach response begins with a record count and a notification assessment. How many individuals, in which jurisdictions, under which statute. ‍ Some organizations hold almost no personal data and enormous quantities of other companies' commercial confidences. An insurer's claims files contain policyholders' loss histories, control failures and settlement amounts. The statutory machinery may not engage at all, and what engages instead is a contract portfolio. ‍

The headline change in Europe's network security directive is scope. Directive (EU) 2022/2555 reaches an estimated hundred and sixty thousand entities across eighteen sectors, roughly ten times what its predecessor covered. ‍ The more consequential change is who decides. Under the previous regime a member state identified operators of essential services individually, through an assessment of criticality and dependency.

Sending privileged material through an external AI service is a disclosure to a third party, and voluntary disclosure to a third party waives privilege. The reasoning is straightforward and a federal court has now applied it. ‍ A second federal court reached the opposite conclusion on the same question within days, on a distinction the first did not draw. The position is genuinely unsettled, and the parts that are settled point at configuration choices rather than at a prohibition. ‍

Almost every obligation in the AI Act is keyed to a risk classification. Work out which tier a system falls into and the duties follow. ‍ Article 4 is not. It applies to providers and deployers of any AI system whatever it does, its subject is people rather than systems, and national authorities began supervising and enforcing it from 2 August 2026 while the high-risk regime moved to December 2027. It is the live one, and it sits outside the structure most governance programs are built on. ‍