AI Threats

How IsAIThreat tracks AI threats around the world

Our method: which sources we monitor, how stories are classified into nine threat categories, and how severity and location are estimated.

Artificial intelligence is now part of fraud, cybercrime, propaganda and warfare, and the reporting on it is scattered across hundreds of outlets. IsAIThreat brings that reporting into one place so readers can see what is happening, where, and how often.

What counts as an “AI threat”

We include a story when it describes harm, misuse, risk or a failure involving an AI system, or a legal or regulatory response to one. Every story is placed in at least one of nine categories:

  • Deepfakes & synthetic media: fake video, cloned voices, face swaps and non-consensual imagery.
  • AI scams & fraud: phishing, impersonation and financial fraud that use AI.
  • AI-powered cyberattacks: malware, ransomware and intrusion campaigns that use or target AI.
  • Jailbreaks & model security: prompt injection, jailbreaks, data poisoning and attacks on AI agents.
  • Privacy & surveillance: data leaks, facial recognition and misuse of personal data.
  • Misinformation & influence operations: AI-generated propaganda and election interference.
  • AI safety & alignment: dangerous capabilities, chatbot harms and frontier-model risks.
  • Regulation & lawsuits: new laws, enforcement actions and court cases.
  • Military & autonomous weapons: AI in conflict and national security.

Where the stories come from

Every few hours we check RSS feeds from established security publications, technology newsrooms, and targeted news searches that cover thousands of outlets worldwide. We publish only the headline, a short excerpt and a link. The full story always stays with the original publisher, and we encourage you to read it there.

A big story is often reported by dozens of outlets within a day or two. Rather than list the same event again and again, the feed shows it once and lists the other outlets’ reports underneath. The grouping is automatic and errs on the side of keeping stories apart, so two reports of one event occasionally still appear separately. The counts and charts on the tracker include every report, because the number of outlets covering an event is itself a sign of how much attention it is getting.

Severity and location are estimates

Each story gets an automatic severity estimate:

  • High: threats to life, critical infrastructure, elections, children, or attacks that are actively exploited or state-sponsored.
  • Medium: confirmed malware, breaches, fraud, deepfakes or lawsuits.
  • Low: research, policy discussion and warnings.

Countries and regions come from places named in the headline and excerpt. A story with no clear location is marked Global / Unspecified.

These labels come from keyword rules, not human review, so some will be wrong. Treat them as a guide to the volume and spread of reporting, not as verified incident data.

Why this matters

Single headlines are easy to dismiss. Seen together over weeks and months, they show trends: which threats are growing, which regions are being targeted, and whether regulation is keeping up. The Threat Tracker charts those trends, and this Analysis section explains the most important ones.

Have a correction or a story we missed? Contact us.