The NATO Strategic Communications Centre of Excellence (NATO StratCom COE) exists to help the Alliance understand and counter the information threats shaping modern conflict, from state-sponsored propaganda to the newest generation of AI-enabled influence campaigns. As large language models became powerful enough to generate convincing, targeted content at scale, StratCom COE needed more than isolated research. It needed a shared operational model that every NATO member could use to detect and disrupt AI-driven disinformation before it spread. That is the work StratCom COE undertook with Enkrypt AI, an AI security and compliance company now part of Anaconda.
The Challenge: A Threat That Outran Fragmented, National Defenses
Before this joint research, NATO members largely worked from their own vocabulary, toolchains, and timelines when studying AI-enabled influence operations. Detection happened at the point of viral spread, after the damage was already done. Cross-platform and cross-language signals were rarely correlated until an operation had already shaped public opinion.
Meanwhile, the underlying technology was advancing quickly. Adversaries no longer need a single powerful model to run a disinformation campaign. They can assemble commercial and open-source models, including models with safety fine-tuning deliberately stripped away, into coordinated systems that discover vulnerable audiences, generate synthetic personas, and craft narratives for deployment across platforms. In related Enkrypt AI research, a simulated disinformation system built this way was shown to feed evaluation results back into earlier stages, continuously self-optimizing faster than human moderators could respond. That same related research, adversarial testing of eight frontier language models, found every model exploitable to some degree, with “abliterated” open-source models, whose safety refusals have been surgically removed from the model weights, posing the most acute risk. Point-in-time detection and single-model safety reviews were not built for a threat that operates at that speed and scale.
The Approach: Mapping the Full Lifecycle, Not Just the Symptoms
NATO StratCom COE partnered with Enkrypt AI to move from theory to an operational framework. Rather than studying isolated incidents, the two teams modeled how an AI-enabled influence operation runs end to end: target discovery and audience segmentation, persona generation, narrative generation, coordination patterns, and cross-platform relay. Evaluation results from each stage are fed back into the model, so the framework captures how these campaigns learn and improve over time, not just how they look at a single moment.
The research paired Enkrypt AI’s adversarial red-teaming discipline, the same discipline it applies across its security platform, with alliance-level analysis of how AI-generated content actually spreads. That combination let the team catalogue distinct influence tactics, analyze engagement signals across coordinated activity, and trace how many real users organically echoed campaign narratives: the clearest evidence that a synthetic campaign had achieved genuine downstream influence.
The result reframed a diffuse, fast-moving risk as a structured model NATO members could apply consistently, built around six pillars: multi-persona modeling, a full lifecycle map, a tactic catalogue, engagement analysis, adoption tracing, and an operational framework for detection and defense.
The Results: From Reactive Research to a Shared, Reusable Defense
Applying that six-pillar framework across the partnership translated directly into measurable outcomes:
- 47 distinct influence tactics mapped, documenting how each is created, scaled, and deployed across platforms
- 60,000+ engagement signals analyzed to surface personas, narrative convergence, and amplification patterns
- 350+ organic narrative adoptions traced, real users independently echoing campaign talking points, validating that the modeled threat produces measurable real-world influence
Detection cadence moved from post-hoc to upstream. Shared terminology across the Alliance went from partial to complete. And instead of each member re-solving the same problem in isolation, the framework was built for reuse by design, giving every NATO mission a common model and vocabulary for a threat that does not respect borders.
“Enkrypt AI translated a fast-moving, theoretical threat into an operational model we can use to better design defense strategies against automated influence threats. Their work gave us a practical, technically informed framework for identifying AI-enabled influence campaigns and reinforcing our collective information resilience.”
Joshua Chia, Senior Expert, NATO Strategic Communications Center of Excellence
Looking Forward
The framework NATO StratCom COE and Enkrypt AI built is designed to extend as the threat evolves: adaptable to new platforms, new model families, and new tactics as adversaries develop them. For an alliance built on shared defense, that reusability matters as much as the initial findings. A detection blueprint one member validates becomes an asset every member can apply.
Enkrypt AI is now part of Anaconda, bringing the same adversarial-testing rigor and policy-mapped risk assessment that powered this research into Anaconda’s broader mission of helping enterprises and governments secure, orchestrate, and accelerate their AI systems at scale. As AI models and the threats built around them keep advancing, red-teaming depth and operational clarity is exactly what defenders need to stay ahead, whether they are running an enterprise SOC or a NATO mission.
Ready to red-team your AI systems before adversaries do? Contact our team to learn how we can help you build a proactive AI defense.
AI Security and Guardrails
Protect them at runtime, govern tool and data access, and continuously prove compliance. Security Built into your AI, not bolted on after.