Open societies have a paradox at their core. The freedoms that make democratic governance the most durable, innovative, and humane system of political organization ever devised — freedom of expression, open scientific inquiry, transparent markets, the rule of law — are the same freedoms that produce the technologies our adversaries study, steal, and weaponize against us.

This is not a new problem. It predates AI by decades. What is new is the scale, speed, and precision with which authoritarian states can now deploy our own tools against us. Agentic AI — networks of coordinated AI agents that reason, plan, and act autonomously with persistent memory and cross-domain reach — has crossed a threshold that changes the calculus.

The United States, its allies, and the open societies they represent need to understand what is happening, who is doing it, how they are doing it, and what to watch for. This is not a classified briefing. It is what informed citizens and decision-makers should know right now.

Why Open Societies Produce the Tools That Threaten Them

The irony is structural. Democracies invest in open research because transparency produces better science. They publish findings because peer review improves quality. They release open-source software because distributed development accelerates capability. OpenClaw, the multi-agent orchestration platform that powers sophisticated AI operations, is open source. The frontier models that drive it — built by American, British, and allied researchers — are accessible via API to anyone with a credit card.

China does not have a comparable culture of open publication in strategic technology domains. Russia does not. Iran and North Korea do not. But they have researchers who read Western journals, engineers who clone Western repositories, and intelligence services that have spent decades acquiring what they cannot build.

The governance layer is where the asymmetry becomes dangerous. When a democratic organization deploys an agentic AI system, it builds in escalation protocols, human oversight requirements, audit trails, and ethical guardrails. These are features of accountable governance. When an authoritarian state deploys the same architecture, it removes those constraints deliberately. The result is a system with identical capabilities operating at higher speed, higher scale, and with no friction from conscience or accountability.

Five Ways Adversaries Are Using Agentic AI Against Us

1. Influence Operations at Industrial Scale

The clearest and most documented threat is AI-accelerated influence operations. China’s GoLaxy network, exposed by Axios in August 2025, used DeepSeek’s open-source reasoning model to create synthetic personas capable of adapting their messaging to specific audiences and mimicking real people. This is agentic behavior — autonomous, adaptive, persistent, personalized.

The World Economic Forum warned in March 2026 that deepfakes have crossed a critical threshold: they have eliminated the tell-tale glitches that once made them detectable, and are now accessible to anyone with a smartphone. The combination of undetectable synthetic media and autonomous persona networks represents a qualitative shift in the influence operation threat.

What this means in practice: adversary agents can now operate thousands of synthetic social media accounts simultaneously, each with a coherent persona, posting history, and adaptive messaging strategy. They can target specific demographics, geographies, and psychological profiles. They can respond to breaking news faster than human operators. And they can do all of this at near-zero marginal cost.

2. Cyber Offense: AI-Accelerated Attack Cycles

Every phase of a cyberattack — reconnaissance, vulnerability identification, exploit development, lateral movement, exfiltration — can be accelerated by AI. Adversaries are not waiting for permission to use these capabilities.

North Korea’s Lazarus Group and affiliated threat actors have long used cyber operations to fund the regime and steal intellectual property. Their operations against the cryptocurrency sector have netted billions. AI-assisted reconnaissance and spearphishing dramatically reduce the time and human capital required to execute sophisticated intrusions. An agentic system can identify targets, craft personalized lures, monitor for successful compromises, and escalate to human operators — all without a person in the loop until the intrusion is already established.

Iranian threat actors have demonstrated similar evolution. The Islamic Revolutionary Guard Corps’ cyber units have moved from blunt denial-of-service attacks to sophisticated, patient intrusion campaigns targeting critical infrastructure. AI accelerates the patient phase — the reconnaissance, the mapping, the waiting.

3. Economic Espionage: Systematic IP Theft at Scale

China’s strategic technology acquisition program is the largest intellectual property theft operation in human history. AI makes it faster and harder to detect. Agentic systems can monitor patent filings, academic preprints, corporate job postings, and conference proceedings continuously — building a real-time picture of where American and allied technology is advancing and where the gaps are worth exploiting.

Insider threat identification, supply chain infiltration, and academic collaboration exploitation are all enhanced by AI systems that can process vast quantities of open-source data and identify patterns that human analysts would miss. The target is not just secrets — it is the entire frontier of Western innovation.

4. Military Planning and Wargaming

Authoritarian militaries are using AI for the same applications democratic militaries are exploring: logistics optimization, targeting support, intelligence fusion, and wargaming. The difference is oversight. A Chinese or Russian AI-assisted targeting system does not have a judge advocate general reviewing outputs for compliance with the laws of armed conflict.

The EU DisinfoLab December 2025 report documents how state-backed influence operations are increasingly integrated with broader hybrid warfare strategies — information operations synchronized with cyber operations, economic coercion, and military signaling. Agentic AI enables tighter coordination across these domains than was previously possible.

5. Domestic Repression Exported

This is the threat that receives the least attention in Western security discourse and deserves more. China’s domestic AI surveillance architecture — social credit systems, predictive policing, facial recognition at scale, mass monitoring of communications — is not staying inside China’s borders. It is being exported, sold, and deployed in authoritarian and semi-authoritarian states across Africa, Southeast Asia, the Middle East, and Latin America.

The technology that suppresses dissent in Xinjiang today is being exported to suppress dissent in Venezuela, Ethiopia, and Cambodia tomorrow. Agentic AI makes these systems more capable, more autonomous, and more scalable. The global expansion of AI-enabled authoritarianism is a direct threat to the conditions that make free societies possible.

What Every Informed Person Should Know and Watch For

At the societal level:

  • Coordinated inauthentic behavior at scale — social media accounts that post at inhuman frequency, adapt messaging rapidly, and cluster around divisive topics — is increasingly AI-driven. When you see it, report it. Don’t amplify it.
  • Deepfakes are no longer detectable by eye. Verify before sharing. Demand provenance. If a video of a public figure saying something explosive appears from an unverified source, treat it as suspect until confirmed by multiple credible outlets.
  • The U.S. government has been pulling back resources dedicated to countering foreign disinformation. That means civil society, journalism, and informed citizens carry more of this burden than they did five years ago.

At the organizational level:

  • Spearphishing has become hyper-personalized. An AI system can craft a phishing email that references your actual colleagues, your actual projects, and your actual communication style. Train your people to verify unexpected requests through out-of-band channels regardless of how authentic they appear.
  • Third-party software and supply chain compromise is a primary adversary vector. Know what is in your software stack. Apply patches. Audit access.
  • If your organization works in defense technology, critical infrastructure, finance, or advanced research, assume you are a target. Act accordingly.

At the policy level:

  • Governance frameworks for agentic AI — escalation protocols, audit requirements, human oversight mandates — are not bureaucratic friction. They are the structural difference between a tool that serves democratic values and one that can be weaponized without accountability. Defend them.
  • Export controls on AI chips, models, and agentic platforms matter. The technology transfer debate is not abstract — it directly affects how quickly adversaries can close the capability gap.
  • Investment in open-source intelligence, attribution, and adversary AI monitoring is a national security requirement, not a nice-to-have.

The Structural Advantage We Must Not Squander

Open societies are not defenseless. We built these tools. We understand them better than our adversaries do. We have the talent, the infrastructure, and — if we choose to exercise it — the institutional will to deploy AI in defense of the values that produced it.

The governance gap cuts both ways. Yes, adversaries can move faster without oversight. But systems without oversight are also systems without accountability, correction mechanisms, or the trust of the populations they serve. Authoritarian AI is brittle in ways that democratic AI, built with transparency and accountability, is not.

The answer is not to remove our guardrails to match adversary speed. It is to build faster within them, to invest in detection and attribution, and to ensure that the benefits of agentic AI — the same productivity, intelligence, and decision-support advantages that authoritarian states are pursuing — are distributed broadly across the free world first.

The tools are ours. The values are ours. The urgency is now.

References