Within the next two months, at least one publicly disclosed AI supply-chain security incident will attribute a compromise to a malicious model, hidden backdoor or poisoned weights hosted on a major open model repository (Hugging Face, GitHub, or a hyperscaler model hub).

Reasoning trace

1. Rapid growth of open-source model consumption creates a large unaudited attack surface
2. Signal topic on AI model security vulnerabilities (398 signals, new) indicates active research disclosure
3. Researchers or attackers publicly demonstrate real-world exploitation via a poisoned weight or dependency
4. A vendor or CERT issues a formal advisory naming the repository

Evidence so far

24 confirming · 0 denying signals · 8 verified · 1 broken assumptions. The evidence itself is part of the full dossier.

Probability history

Updated 29 times since mint (last on 2026-10-01) — 34% at mint → 84% today.

What to watch

• milestone Security researchers or threat actors publicly disclose technical details of the compromise (e.g., via a GitHub issue, blog post, or conference presentation) identifying the specific poisoned model or backdoor observed
• milestone A security researcher or organization publishes a technical blog post, GitHub repository, or security bulletin detailing the discovery of the backdoored model and providing reproducible proof of compromise
• leading indicator The affected model repository (Hugging Face, GitHub, or hyperscaler) issues a public notice or takedown regarding the compromised model or initiates a security investigation