AI disinformation campaigns: a ban is not enough

On October 8, Anthropic published the annual update to its Usage Policy, which takes effect November 12. Most of the changes clarify rules that already existed. One of them goes straight to the heart of reputation work: the company has pulled into a single section what used to be scattered across its elections, fraud, privacy, and disinformation rules. Its models may not be used to obscure who is behind a message, to amplify it through fake accounts or posts, or to build the tools those campaigns run on. It applies to political and commercial deception alike.

This is not a theoretical ban. The company itself says it has seen state media outlets, government propaganda offices, and commercial firms use its model to run networks of fake accounts and fabricated news sites.

I think it is a sound and necessary measure. I also think it is nowhere near enough, and that an organization that feels safer because of it is making a mistake.

Attackers do not ask for permission

A usage policy binds the people who use the services of the company that wrote it. It keeps a serious company from lending its technology to these campaigns, and that counts for something: it closes the most convenient door.

But the people who do this for a living do not depend on that door. There are open models anyone can download and run on their own machines, and there are versions of those models whose built-in limits someone has removed; in the field they are called “abliterated.” They come with no terms of use to follow and nobody watching what they are asked to do. They can do the same damage as a commercial model, or more.

So the policy governs the behavior of those who already meant to comply. For someone who has decided to attack, it changes the tool, not the intent.

What has changed, and what has not

I have seen disinformation and smear campaigns since the days when they were run by hand or with very basic scripts. Later came more elaborate ones, with photo and video editing and with subcontracted writers, in Asia for English and in Latin America for Spanish. They took people, coordination, and budget, and that limited who could commission them and against whom it was worth the trouble.

The underlying technique is the same as it was then. What has changed is that today it is within reach of many, it costs less, and above all it can be automated. Computer security has a comparison that explains it well. A denial-of-service attack, a DoS, floods a server from a single machine: it is a nuisance, but it can be traced and cut off. A DDoS does the same from thousands of machines at once, and there is no longer one source to shut down. That is what has happened to disinformation campaigns: from DoS to DDoS.

The consequence is uncomfortable. Destroying a reputation has always been easier and cheaper than building one, and now it is more so. You no longer need to be a big target for someone to find it worth their while.

What an organization can do

The first thing is to analyze its environment and watch its entire digital ecosystem: what is being said, where, who is pushing it, and since when. With beacons and sonars that raise the alarm when something moves, before it reaches the first page of a search engine or the answer of an assistant. That is the work we do at Kuantia.

The second matters far more, and it is what gets done least: training communications teams to act and to decide. Knowing that something is happening is of no use if nobody knows what to do about it, who decides, and how fast.

The worst mistake in communications is not communicating. Directly or indirectly, but you have to be there. Whoever gives up the space to others loses it, and with it their voice and their weight in the message. Winning it back later costs far more than holding it in time.

Before, not during

The solution does not lie in the developers’ policies, though they are right to write them. It lies in detoxifying your own digital horizon and in arriving at a reputation crisis prepared: with the environment watched, the team trained, and the decisions thought through in advance.

That is the framework we work in with our clients. Si vis pacem, para bellum.

The linked source was consulted on October 9, 2026.

Blog