AI Recommendation Poisoning in "Summarize with AI" buttons
Technical Analysis
Summary
Hide ▲
Show ▼
AI Recommendation Poisoning now shows how clickable "Summarize with AI" buttons can inject persistent prompts into chatbot memory, creating a new path to bias recommendations and citations. The technique matters because it can silently steer outputs toward a chosen source across health, finance, and security topics. It also broadens the abuse surface by turning ordinary links and embedded buttons into a delivery mechanism for URL-based memory manipulation.
Related Happenings
LayerX font-rendering PoC exposes a browser-rendering gap in AI assistant analysis
Technical Analysis
First: 17.03.2026 15:59
Last: 17.03.2026 15:59
Sources 1
About this happening:
A **LayerX** proof-of-concept showed that a **font-rendering attack** can hide malicious webpage commands from AI assistants, creating a risk of **unsafe guidance** when the brows...
LayerX font-rendering PoC exposes a browser-rendering gap in AI assistant analysis
Technical AnalysisAbout this happening: A **LayerX** proof-of-concept showed that a **font-rendering attack** can hide malicious webpage commands from AI assistants, creating a risk of **unsafe guidance** when the brows...
AI assistants with web browsing repurposed as covert C2 relays
Technical Analysis
First: 18.02.2026 17:00
Last: 18.02.2026 17:00
Sources 1
About this happening:
AI assistants with web browsing are now being shown as **covert command-and-control relays**, letting malware hide commands and stolen data inside routine enterprise traffic. **Gr...
AI assistants with web browsing repurposed as covert C2 relays
Technical AnalysisAbout this happening: AI assistants with web browsing are now being shown as **covert command-and-control relays**, letting malware hide commands and stolen data inside routine enterprise traffic. **Gr...
AI as a C2 proxy abuse of Microsoft Copilot and xAI Grok browsing channels
Technical Analysis
First: 17.02.2026 20:08
Last: 17.02.2026 20:08
Sources 1
About this happening:
Researchers disclosed **AI as a C2 proxy**, a technique that can turn **Microsoft Copilot** and **xAI Grok** browsing features into stealthy **command-and-control relays**, increa...
AI as a C2 proxy abuse of Microsoft Copilot and xAI Grok browsing channels
Technical AnalysisAbout this happening: Researchers disclosed **AI as a C2 proxy**, a technique that can turn **Microsoft Copilot** and **xAI Grok** browsing features into stealthy **command-and-control relays**, increa...
Tenable One AI Exposure launches as a cloud-native add-on for AI exposure management
Security Tool/Service
First: 30.01.2026 22:23
Last: 30.01.2026 22:23
Sources 1
About this happening:
**Tenable** has launched **Tenable One AI Exposure**, a cloud-native add-on that helps enterprises detect and govern **agentic and generative AI** use before it creates **data exp...
Tenable One AI Exposure launches as a cloud-native add-on for AI exposure management
Security Tool/ServiceAbout this happening: **Tenable** has launched **Tenable One AI Exposure**, a cloud-native add-on that helps enterprises detect and govern **agentic and generative AI** use before it creates **data exp...
Timeline
-
17.02.2026 11:31 2 articles · 3mo ago
Microsoft discloses AI Recommendation Poisoning
Initial DisclosureMicrosoft publishes research on AI Recommendation Poisoning in "Summarize with AI" buttons, showing that hidden instructions in URL prompt parameters such as "?q=" can inject persistence commands into chatbot memory and bias future recommendations. The findings describe more than 50 unique prompts from 31 companies across 14 industries over a 60-day period, note that crafted links can be delivered through web pages and email, and identify turnkey tooling such as CiteMET and AI Share Button URL Creator that lowers the barrier to deploying manipulative AI links.
Show sources
- Microsoft Finds “Summarize with AI” Prompts Manipulating Chatbot Recommendations — thehackernews.com — 17.02.2026 11:31
- Microsoft Finds “Summarize with AI” Prompts Manipulating Chatbot Recommendations — thehackernews.com — 17.02.2026 11:31