Artificial intelligence is being used as a tool by people who want to defraud, attack or mislead others. Some of that has a location and shows up on the Situara map. Some of it has no location at all, and belongs on a page like this one instead. Both are here.
Updated weekly. Open the live map and switch on the AI & Cyber layer to see these in place.
A video or voice recording of a real person, generated by software, used to get money or access. The well-known form is a call or video meeting in which a colleague, a director or a family member appears to ask for an urgent payment. The face and the voice are convincing; the request is not.
What you can do
Attackers use AI to write phishing messages without the language errors that used to give them away, to adapt malware, or to run large numbers of break-in attempts automatically. The attack itself is not new. The speed, the volume and the quality of the bait are.
What you can do
Generated text, images, audio and video used to influence opinion, often around elections or a conflict, and often attributed to actors outside the country being targeted. What makes it different from older propaganda is cost: a campaign that once needed a staff of people can now be run by a few.
What you can do
Jamming drowns out the satellite signal; spoofing replaces it with a false one, so a ship or aircraft is told it is somewhere it is not. This has been going on for years in and around conflict zones. Where a report explicitly says AI or autonomous systems are part of it, we mark the report on the existing GNSS Disruption layer. We do not assume it.
What you can do
Not every AI risk can be put on a map, and pretending otherwise would make the map worse. The list below is the wider picture, ordered by how firmly it is established today. It is background, not a live feed.
| Status | Risk | What it means in practice |
|---|---|---|
| Happening | Deepfakes and voice cloning | Identity is no longer proven by a face or a voice. Affects fraud, evidence and trust between people. |
| Happening | Automated cybercrime | Phishing and malware produced faster, cheaper and in better language than before. |
| Happening | Disinformation at scale | Influence campaigns that used to need a newsroom can be run by a handful of people. |
| Happening | Bias in automated decisions | Systems that decide on credit, work, benefits or policing can repeat and harden existing unfairness. |
| Happening | Privacy and data exploitation | Personal material is collected and combined to train systems, usually without a real choice. |
| Happening | Work and income | Tasks disappear faster than the people doing them can move to other work. |
| Emerging | Concentration of power | A small number of companies control the models much of the economy starts to rely on. |
| Emerging | Single-model dependency | When many services build on the same model, one fault or outage hits all of them at once. |
| Emerging | Autonomous weapons | Targeting decisions moving from a person to a system, with accountability unresolved. |
| Emerging | Loss of skill | Professions that stop practising a judgement they still need when the tool is wrong or unavailable. |
| Debated | Loss of control | Systems pursuing goals that differ from what was intended, in ways that are hard to correct. |
| Debated | Existential risk | Argued seriously by part of the research community and disputed by another part. Unresolved. |
Once a week we read a set of news sources and let a language model pull out only what is literally stated in them. Each record keeps the article it came from, and that article is the link in the map popup. Four rules keep it honest:
The result is a thin layer, usually a handful of reports. That is deliberate. A map people use to judge whether somewhere is safe is worth less, not more, when it is filled with items that only might be true.
This layer is free for everyone, like the wildfire layer. See what the map layers mean for the rest of the map, or where our data comes from.