AI Assistants Exploited via Crafted Links for Recommendation Poisoning

# AI Assistants Exploited via Crafted Links for Recommendation Poisoning

https://gcdradar.vercel.app/incidents/ai-assistants-exploited-via-crafted-links-for-recommendation · AIR-2026-0112 · closed 2026-09-29

- Lane: Patch now
- Severity: 3 (Confirmed exploitation)
- Fix status: Fix status unknown
- First reported: 2026-09-15 (happened 2026-02-10)
- AI role: Attack on AI
- Products: AI Assistants

## Summary

Operators embedded malicious links into websites to hijack initial user turns with AI assistants. The attack manipulated assistant memory and recommendation outputs for promotional purposes.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-09-15: Manipulating AI memory for profit: The rise of AI Recommendation Poisoning (https://microsoft.com/en-us/security/blog/2026/02/10/ai-recommendation-poisoning)
- 2026-09-15: AI Recommendation Poisoning via Crafted AI Assistant Links (https://atlas.mitre.org/studies/AML.CS0072)

## Sources

- 2026-09-15 MITRE ATLAS: AI Recommendation Poisoning via Crafted AI Assistant Links https://atlas.mitre.org/studies/AML.CS0072
- 2026-09-15 Microsoft Security: Manipulating AI memory for profit: The rise of AI Recommendation Poisoning https://microsoft.com/en-us/security/blog/2026/02/10/ai-recommendation-poisoning

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.