IN Brief:
- Flashpoint says illicit AI-related activity surged late in 2025, with compromised credentials, ransomware, and vulnerability exploitation converging into faster attack cycles.
- For defence manufacturers, the exposure runs through identities, software pipelines, supplier access, and connected production systems rather than the old perimeter alone.
- The result is a harder industrial requirement: cyber resilience now has to be designed into engineering, factory operations, vendor management, and product support.
Flashpoint’s latest global threat reporting points to a shift that defence manufacturers can no longer treat as a security team problem sitting somewhere outside operations. The issue is not simply that artificial intelligence is showing up in cybercrime. It is that agentic AI is helping attackers automate reconnaissance, iterate attacks more cheaply, rotate infrastructure, and learn from failure at a pace that compresses the time available for defenders to notice and respond.
That matters sharply for the defence industrial base because the preferred entry points are often not the heavily protected crown-jewel systems people talk about first. They are the contractors, vendors, software environments, and adjacent platforms that sit around government and prime networks. Flashpoint-linked reporting says some of this activity has been associated with Chinese advanced persistent threat groups targeting platforms widely used by government agencies, multinational companies, and defence contractors.
The broader metrics are ugly enough on their own. Flashpoint says AI-related illicit activity jumped steeply at the end of 2025, while stolen credentials and cloud tokens have become central to how attackers gain access. The report also points to rising ransomware activity and a heavier vulnerability burden, which together reinforce a simple point: attack chains are becoming more industrialised, more automated, and less dependent on patient manual tradecraft.
For defence manufacturers, that changes the risk map. A factory making avionics, power electronics, vehicle subsystems, or guidance hardware may not be breached through the production line itself. It may instead be reached through identity systems, remote vendor access, engineering collaboration platforms, update servers, or compromised cloud sessions that allow an attacker to behave like a legitimate user. When that happens, the boundary between cyber espionage and production disruption gets very thin.
Why factories sit in the blast radius
Modern defence production is dense with digital dependencies. Design data flows between engineering tools and programme partners. Test equipment depends on networked software. Manufacturing execution systems, quality databases, patching workflows, and supplier portals all create connective tissue that did not exist in older plant models. Even organisations with segmented operational technology often retain awkward bridges between business systems, lab environments, and factory-floor support functions.
That is why static perimeter logic is ageing badly. An attacker armed with valid credentials, a foothold in a software pipeline, or visibility of a vendor relationship does not need to smash through the front gate in theatrical fashion. They can move through the ordinary mechanisms that keep modern production running. In a defence context, the objectives may include design data, schedules, tooling parameters, support documentation, or confidence in the integrity of what leaves the line.
Government cyber bodies have already been nudging industry in this direction. NSA’s DIB support work is built on the premise that contractors are frequently targeted by nation-state actors, while DoD’s manufacturing overlay makes the same point more dryly: manufacturing systems require tailored controls, not generic office IT assumptions. That is the operational reality for any supplier whose plant environment now depends on software, sensors, and connected equipment.
Cyber resilience becomes a production discipline
The response therefore has to be industrial, not cosmetic. Identity governance matters more. Asset visibility matters more. Supplier access controls matter more. So do software provenance, privileged account monitoring, segregation between engineering and production environments, and recovery plans that assume an attacker may already have credentials rather than merely malicious code.
This is also becoming a board-level manufacturing issue because output is now part of the target set. A cyber incident that corrupts a build environment, delays acceptance testing, freezes design collaboration, or undermines trust in quality data can hit delivery schedules just as effectively as a physical supply interruption. The factory does not need to go dark to become commercially and operationally compromised.
Agentic AI sharpens that problem by reducing the attacker’s cost of experimentation. Failed attempts are cheaper, phishing lures improve faster, and weak points in sprawling supplier ecosystems become easier to probe continuously. For the defence industrial base, the implication is plain enough: cybersecurity can no longer be bolted onto manufacturing after the line is designed. It now sits inside the production model itself.
