Defense-industrial access can form before a conventional security workflow has a stable employee, device, or account to monitor. DPRK IT-worker schemes, suspected Iran-nexus recruitment operations, and supplier compromise remain distinct threats, but they expose the same defensive problem: workforce and delegated authority are often created from evidence scattered across teams that do not routinely compare records.

Device-code phishing turns a legitimate login flow into rented access. The durable defense is to shrink who can use it.

A lot of KEV response still collapses into one closure condition: the vulnerable system was patched. That works if exploitation never happened. It is weaker when the asset was exposed, telemetry is incomplete, or the system carries identity, remote-access, payment, or administrative authority.

AI gateways are starting to concentrate credentials, logs, routing, quotas, and policy. That makes them worth watching now.

World Cup fraud shows why removing infrastructure is not the same as disrupting the operation.

MCP is not just an AI security story. It may be the first real test of agent connector supply-chain risk.

Your agent kept notes. AI-agent memory is not vibes. It is storage.