What Happened
Unit 42 reports a cyberespionage campaign targeting Southeast Asian governments that deployed the PerfectPioneer remote-access trojan and modular KUKURAT malware. The attackers used HTTP/3 for stealthy command-and-control communications. The report does not establish that AI systems were targeted or that AI was used in the campaign [1].
Why It Matters to Businesses
AI applications depend on endpoints, cloud workloads and outbound network access. A compromised host running an AI service could use those same paths to contact an attacker, regardless of whether the application itself has an AI-specific vulnerability. Security teams that inspect web traffic primarily through HTTP/1.1 or HTTP/2 controls should verify what visibility they retain when traffic uses HTTP/3 [1].
Kimbodo Engineering Perspective
Do not mistake a transport technique for an AI-specific attack. The useful lesson is architectural: protecting model calls and agent tools is not enough if a compromised workload can establish unmonitored outbound connections. Network controls should be paired with endpoint detection and tightly scoped workload identities; neither layer is sufficient alone.
How We Would Implement It
- Inventory which AI services and supporting hosts need outbound HTTP/3, and define approved destinations for each workload.
- Route production AI workloads through controlled egress. Restrict outbound UDP/443 where HTTP/3 is unnecessary, and test application behavior before enforcing the policy.
- Correlate DNS, network-flow and endpoint telemetry to investigate unexpected destinations, persistent connections and suspicious processes. Do not rely on payload inspection where encrypted traffic is unavailable to inspect.
- Run agents and tool-execution workers with least-privilege credentials and segmented network access, so a compromised process has fewer paths to sensitive data or external systems.
Risks, Costs and Security
Blocking UDP/443 can disrupt legitimate applications or cause them to fall back to another protocol; it does not, by itself, prevent command-and-control traffic. Egress controls require destination maintenance, while endpoint telemetry adds operational cost and may not be available on every managed service. Prioritize coverage of workloads with sensitive data access, then validate the controls with authorized tests of HTTP/3 visibility and incident response.
Where Kimbodo Comes In
Kimbodo builds and operates this in production for businesses — see our AI Security & Guardrails practice, or Request a Security Review.