In September 2026, a threshold was crossed in the long history of states shaping foreign minds: Iran and China deployed influence campaigns run not by human operators, but by autonomous artificial intelligence systems capable of generating, distributing, and refining persuasive content without rest or hesitation. Built on Chinese open-source models freely available to the world, these operations revealed that the most consequential innovations in information warfare no longer require secrecy — only ingenuity in weaponization. The discovery forces a reckoning not merely with what these particul
Iran and China Deploy First Autonomous AI Influence Campaigns
The barrier to entry for state actors was lower than ever.
Mark
So these are not human operators running fake accounts anymore. The AI is actually making decisions about what to post?
Mimi
Exactly. The autonomous agents generate content, choose when and where to deploy it, and adjust based on how audiences respond. It's continuous and it doesn't get tired.
Luke
But we should be careful here—the reporting says the campaigns used these systems, but I don't see specifics about how much of the operation was actually autonomous versus how much still required human direction. That matters.
Mark
Why does it matter?
Luke
Because "autonomous" can mean different things. If a human still has to approve each post, that's one thing. If the AI is truly making independent decisions about messaging strategy, that's another. The source material doesn't distinguish.
Mimi
Fair point. What we know for certain is that Iranian and Chinese actors combined open-source AI models with agent frameworks. The scale and coordination were new. Whether every decision was autonomous or some were human-guided, the capability is still unprecedented.
Mark
And Israeli firms were involved. That's surprising.
Luke
It is, but the reporting doesn't explain how. Were they knowingly participating? Did their tools get repurposed? That's a crucial gap.
Mimi
The reporting acknowledges the ambiguity. It says their involvement "suggested a blurred line" but notes that whether they knew what they were contributing to "remained unclear."
Mark
So what happens next? Can platforms actually stop this?
Luke
That's the real question, and the reporting is honest that nobody has a clear answer yet. Traditional defenses—removing bot networks, flagging coordinated behavior—might not work if the AI is operating within platform rules.
Mimi
Which is why this moment feels like a threshold. The technology existed in theory. Now it's been deployed. The questions about governance are no longer academic.
Le Pouls
- For the first time, state-sponsored influence campaigns ran almost entirely without human hands — AI agents generating thousands of tailored messages, deploying them across platforms simultaneously, and adapting in real time to what audiences responded to.
- The technical barrier that once made such operations costly and detectable has collapsed: Iran and China built these systems not from secret proprietary tools, but from open-source AI models released freely to the global research community.
- Israeli private firms appeared inside the technical architecture of these campaigns, blurring the line between commercial AI development and active information warfare in ways that initial reporting could not fully untangle.
- Platform moderators and fact-checkers found their conventional defenses — flagging bots, removing inauthentic networks — increasingly ineffective against agents that operated within platform rules and learned faster than human analysts could track.
- Governments and researchers are now confronting questions that were largely theoretical until this moment: how to regulate open-source AI exports, what platforms owe the public in detection, and whether international norms can be built before the technology outruns every agreement.
In September 2026, a threshold was crossed in the long history of states shaping foreign minds: Iran and China deployed influence campaigns run not by human operators, but by autonomous artificial intelligence systems capable of generating, distributing, and refining persuasive content without rest or hesitation. Built on Chinese open-source models freely available to the world, these operations revealed that the most consequential innovations in information warfare no longer require secrecy — only ingenuity in weaponization. The discovery forces a reckoning not merely with what these particular actors have done, but with what any sufficiently motivated actor may now do, and how little the world's defenses were prepared to meet them.
In September 2026, researchers monitoring online influence operations encountered something genuinely new: campaigns coordinated almost entirely by autonomous AI systems, deployed jointly by Iran and China. Unlike the labor-intensive state-sponsored operations of the past — which required human writers, account managers, and coordinators — these systems generated and distributed content across multiple platforms simultaneously, adjusted their messaging based on real-time engagement data, and sustained operations continuously without fatigue.
What distinguished the campaigns technically was not secrecy but ingenuity. Iran and China had not built proprietary systems from scratch. They had taken Chinese open-source AI models — software freely available to the global research community — and integrated them with autonomous agent frameworks. The barrier to entry for this kind of operation had never been lower. The innovation was not in the technology itself, but in weaponizing it at scale.
Researchers also documented the involvement of Israeli private firms in the technical infrastructure of these campaigns, raising unresolved questions about whether those firms knowingly contributed to influence operations or had their tools repurposed without authorization. Either way, their presence illustrated how widely distributed the underlying capability had become — no longer the exclusive province of state intelligence services.
The campaigns flooded platforms with content tailored to local grievances and geopolitical narratives, creating the appearance of organic sentiment while overwhelming human moderators who depended on manual review. Traditional defenses struggled because these were not bots in the conventional sense — they were agents operating within platform rules, learning from their own outputs, and adapting faster than analysts could follow.
The discovery transformed urgent but largely theoretical debates about AI governance into immediate practical crises. How should open-source AI exports be regulated when the models can be weaponized for information warfare? What do platform companies owe the public in detecting autonomous influence operations? Can international norms be established before the technology outpaces any agreement? The campaigns offered no answers — only the pressure of knowing those questions could no longer wait.
In September 2026, researchers tracking online influence operations uncovered something that had not been seen before: coordinated campaigns run almost entirely by autonomous artificial intelligence systems, deployed by Iran and China in tandem. The operations relied on Chinese open-source AI models paired with autonomous agents—software designed to act independently across social platforms, generating and distributing content without human intervention at each step.
The discovery marked a threshold moment in information warfare. Until now, state-sponsored influence campaigns had required human operators to write posts, coordinate timing, manage accounts, and respond to shifting circumstances. Those operations were labor-intensive, visible to pattern-recognition systems, and vulnerable to disruption once discovered. The new campaigns worked differently. The AI agents could generate thousands of pieces of content tailored to specific audiences, deploy them across multiple platforms simultaneously, adjust messaging in real time based on engagement metrics, and sustain operations continuously without fatigue or hesitation.
What made the campaigns particularly significant was their technical architecture. Iran and China had not built proprietary systems from scratch. Instead, they had taken Chinese open-source AI models—freely available software that had been released to the global research community—and integrated them with autonomous agent frameworks. This approach meant the barrier to entry for state actors was lower than ever. The underlying technology was not secret. The innovation lay in weaponizing it at scale and coordinating its deployment across borders.
Israeli firms also participated in these operations, according to the researchers who documented the campaigns. The involvement of private companies alongside state actors suggested a blurred line between commercial AI development and information warfare infrastructure. Whether the Israeli firms knowingly contributed to influence operations, or whether their tools were repurposed without authorization, remained unclear from initial reporting. But their presence in the technical stack underscored how widely distributed the capability had become.
The campaigns targeted multiple audiences across different regions, with content tailored to exploit local grievances, amplify divisions, and shape perceptions of geopolitical events. Because the systems operated autonomously, they could sustain pressure continuously—flooding platforms with coordinated messaging, creating the appearance of grassroots sentiment, and overwhelming human fact-checkers and platform moderators who relied on manual review.
Platform companies and governments had anticipated AI-driven influence operations in theory. But seeing them deployed at this scale and sophistication forced a reckoning with how quickly the technology had matured. The autonomous systems could learn from their own outputs, refine their approach based on what resonated with audiences, and adapt faster than human analysts could track. Traditional defenses—flagging coordinated inauthentic behavior, removing bot networks, labeling state-sponsored content—became harder to execute when the actors were not bots in the conventional sense, but rather AI agents operating within the rules of the platforms themselves.
The discovery raised urgent questions about AI governance that had been largely theoretical until this moment. How should governments regulate the export of open-source AI models if they could be weaponized for information warfare? What obligations did platform companies have to detect and disrupt autonomous influence operations? Could the international community establish norms around the use of AI for influence campaigns, or would the technology simply outpace any agreement? These questions had no clear answers, and the campaigns suggested they would need to be answered quickly.
Envie de l'histoire complète ?
Lire l'original sur The New York Times ↗
War & Conflict