The World Economic Forum confirmed this spring that AI systems were used to optimise political content for maximum emotional impact across multiple countries during the 2024–2025 electoral cycle. Not could be used. Not might be misused. Were used. In actual elections. On real people. With documented effect. Modern AI enables granular segmentation of users for personalised disinformation delivery at industrial scale. Algorithms are now, in the WEF's own language, "autonomous agents in shaping public cognition."

I spent decades in national security watching adversaries develop and deploy influence capabilities that required enormous resources, sophisticated tradecraft, and patient operational preparation. What AI has done is make that tradecraft cheap, fast, and scalable. The barrier to running an influence operation against a major democratic electorate is now within reach of actors who could not have contemplated it a decade ago.

The System That Decides What You See

Most of the public conversation about AI-enabled manipulation focuses on the output: a deepfake video, a fabricated news story, a synthetic persona. These are real threats. But focusing on the output misses the more structurally dangerous problem.

Algorithms do not just distribute content. They construct your information environment. They decide what you see, in what order, framed how, timed to when your cognitive defences are most permeable. They have been optimised over years, with billions of dollars of engineering attention, to maximise engagement. And engagement, at scale, means finding the content most likely to produce a strong emotional response in you specifically.

Your past behaviour. Your stated preferences. Your inferred psychological profile. Your current emotional state, as best the system can detect it from dwell time, scroll speed, and interaction patterns. All of it feeds a model whose purpose is to predict what will make you feel something strongly enough to act.

That is not a bug. It is the product. Your attention, raised to a particular emotional temperature, is what platforms sell to advertisers and, increasingly, to political actors. The architecture was not designed to manipulate you for geopolitical purposes. It was designed to make money. The manipulation is a feature of the optimisation, not a deliberate addition.

Adversaries have simply learned to use the feature.

The Regulatory Response Is Necessary and Insufficient

Two significant regulatory frameworks are converging in 2026. The EU AI Act requires disclosure of AI-generated content at the moment of first exposure, with enforcement beginning in August. A companion Digital Fairness Act takes direct aim at addictive algorithmic design, the mechanism that enables emotional targeting at scale.

Both are meaningful. Neither is sufficient.

Disclosure requirements address synthetic content: if it was AI-generated, say so. That makes sense. But labelling a piece of content does not address the system that decided you should see it today, framed the way it is framed, positioned where you are emotionally susceptible to it. The harder problem is the architecture, not the output.

Sixty-five percent of social media users report wanting more control over their algorithmic experience. That number has been stable across multiple surveys. The gap between what people want and what they have is not an accident. A platform that gave users genuine algorithmic control would be giving up its core business model. The same optimisation that makes platforms valuable to advertisers is what makes them vulnerable to weaponisation.

The Intelligence Frame

In the intelligence world, the most dangerous influence operations are not the ones that fabricate entirely false information. They are the ones that select, sequence, and amplify true information to produce a false understanding. You do not have to lie to manipulate. You have to control the order and the framing.

This is what sophisticated influence operations have always done. What AI has changed is the ability to do it at the individual level, in real time, with continuous feedback. Traditional propaganda was broadcast: the same message to a mass audience, hoping it resonated with enough of them. What algorithmic content optimisation enables is something closer to individual case officers, one for each user, continuously updating their approach based on real-time behavioural data.

The 2024–2025 electoral influence operations documented by the WEF used AI to segment audiences at the granular level and deliver emotionally customised content to each segment. This is not the blunt instrument of Soviet active measures. It is precision targeting, applied to the formation of political beliefs.

Neuroscientist Rafael Yuste, whose work on cognitive liberty has moved from academic to legislative relevance in 2026, has argued that we are approaching a threshold where the distinction between external cognitive influence and internal cognitive process becomes difficult to maintain. The right to think your own thoughts, free from technological intervention you have not consented to, may need explicit legal protection before the technology makes that protection practically impossible to provide.

That argument sounds extreme until you map it against the documented capability of current algorithmic systems. Then it sounds like a reasonable prediction about where the trajectory leads.

The Mind Sovereignty™ Angle

Mind Sovereignty is not an argument for disconnecting from digital information environments. That option is unavailable to most people and increasingly unavailable to all of us. The information environment is the environment we make decisions in. Retreating from it is not a defense.

Mind Sovereignty is an argument for transparency, consent, and architectural accountability. The right to form your own beliefs through your own reasoning, without a system engineered to steer you before you are aware of it, is a precondition of meaningful autonomy. That precondition is under sustained, systematic pressure from systems that are not designed maliciously, but whose incentive structures produce manipulation as a natural output.

The regulatory responses emerging now are necessary. Labelling synthetic content is a reasonable step. Constraining addictive design matters. But the harder conversation, the one about algorithmic accountability and the right to know how the system shaping your information environment is optimised and for whose benefit, is the one we have not yet had at the policy level in any serious way.

The framework that gets there is not content moderation. It is a rights framework that treats cognitive liberty as a fundamental interest worth protecting, and that builds accountability into the architecture of belief formation rather than into the outputs that architecture produces.

That framework does not yet exist in law. The technology it would regulate has been operational for years.

The gap between those two facts is where the war for minds is currently being fought, mostly without rules and mostly without defenders.