Disinformation, fake news and AI fakes

On the page about the information flood we distinguished mis-, dis- and malinformation. Here we look more closely at the targeted variant, and at a new form that makes everything harder: machine-generated content.

Targeted disinformation

Disinformation is invented or distorted and spread on purpose. It pursues a goal: making money, influencing elections, discrediting a group, destroying trust. Typical tools:

• Invented or hijacked sources: websites and accounts that imitate reputable media (similar names, logos, layouts).
• Genuine content torn out of context: an old photo passed off as a current event.
• Coordinated distribution: networks of real and automated accounts (bots) that make a piece of content appear artificially big.

State disinformation

States, too, engage in disinformation. The EU documents pro-Russian campaigns in its database EUvsDisinfo (East StratCom Task Force of the European External Action Service, since 2015), by its own account the largest public collection of pro-Kremlin disinformation, with over 13,000 documented cases (euvsdisinfo.eu).

Example

One documented narrative claimed, without any evidence, that Ukraine was building a „dirty bomb“ at Chernobyl, a fear-inducing story without a source (EEAS). For context: EUvsDisinfo is an EU government source with its own point of view, but individual cases can be traced via the linked original evidence. So even a state fact-checking source is best checked laterally.

The new level: AI-generated content

Artificial intelligence today produces texts, images, voices and videos that look deceptively real. Deepfakes are manipulated or entirely invented media that make real people say or do things that never happened. The basic rule „a photo is proof“ no longer holds.

Example

On 22 May 2023 an apparently AI-generated image of a supposed „explosion at the Pentagon“ spread on Twitter/X, shared among others by verified accounts, including a fake Bloomberg account. The US stock index Dow Jones briefly dropped by around 80 points before it all turned out to be a fake. Verifiers such as Nick Waters of Bellingcat exposed the image by typical AI artefacts: a „floating“ lamppost and a fence that merged unnaturally into the pavement (NPR). A fake with a lifespan of a few minutes, but with real consequences on the stock market.

Recognising AI fakes

There is no 100% method, but there are telltale signs:

• Details that don't add up: hands with too many fingers, unreadable text on signs, distorted backgrounds, „floating“ objects or objects merging into one another.
• Missing confirmation: a dramatic event for which only this one image exists and no reputable source reports is suspicious.
• Check the source: who posted it first? A genuine account or an imitation?
• Check the context: a reverse image search often shows that an image is old or comes from somewhere else (see tools).

A checking routine for viral claims

The News Literacy Project runs RumorGuard, a learning platform that debunks viral false content. It teaches five checking factors that are easy to remember (newslit.org):

1. Authenticity: is the content genuine or manipulated/AI-generated?
2. Source: does it come from a credible source?
3. Evidence: is there evidence that supports the claim?
4. Context: are the place and time correct?
5. Reasoning: is the claim based on sound inference?

In short

With viral content it applies doubly: check first, then believe, and only then share. The more dramatic and emotional, the more important the brief stop.

AI makes faking easier, but the same methods that help against classic disinformation (lateral reading, checking the source, looking for the original) work here too. The technology changes, the basic attitude of checking remains.