Anamorpher open source release for AI image-scaling attack visualization
Security Tool/Service
Summary
Hide ▲
Show ▼
An open source tool named Anamorpher was released to help researchers study image scaling attacks against AI systems. The release matters because hidden prompts can become visible only after preprocessing downscales an image, allowing malicious instructions to reach the model. The tool gives defenders and researchers a practical way to reproduce and analyze this attack path across modern AI products.
Related Happenings
Google Gemini Enterprise and Vertex AI Search security update
Security Patch Release
First: 10.12.2025 14:05
Last: 10.12.2025 14:05
Sources 1
About this happening:
**Google** deployed **updates** to **Gemini Enterprise** and **Vertex AI Search**, reducing a **zero-click** data-leak risk tied to their shared retrieval and indexing workflow. T...
Google Gemini Enterprise and Vertex AI Search security update
Security Patch ReleaseAbout this happening: **Google** deployed **updates** to **Gemini Enterprise** and **Vertex AI Search**, reducing a **zero-click** data-leak risk tied to their shared retrieval and indexing workflow. T...
AI-powered IDEs prompt-injection RCE and data-exfiltration flaws (multiple vulnerabilities)
Vulnerability
First: 06.12.2025 17:24
Last: 06.12.2025 17:24
Sources 1
About this happening:
A disclosed set of **30+ vulnerabilities** in **AI-powered IDEs and coding assistants** creates **data exfiltration** and **remote code execution** risk across tools such as **Cur...
AI-powered IDEs prompt-injection RCE and data-exfiltration flaws (multiple vulnerabilities)
VulnerabilityAbout this happening: A disclosed set of **30+ vulnerabilities** in **AI-powered IDEs and coding assistants** creates **data exfiltration** and **remote code execution** risk across tools such as **Cur...
Google DeepMind CodeMender AI agent debuts to detect and patch vulnerable code
Security Tool/Service
First: 07.10.2025 18:18
Last: 07.10.2025 18:18
Sources 1
About this happening:
Google DeepMind launched **CodeMender**, an **AI-powered agent** that detects, patches, and rewrites vulnerable code to reduce future exploit risk. The system uses **Gemini Deep T...
Google DeepMind CodeMender AI agent debuts to detect and patch vulnerable code
Security Tool/ServiceAbout this happening: Google DeepMind launched **CodeMender**, an **AI-powered agent** that detects, patches, and rewrites vulnerable code to reduce future exploit risk. The system uses **Gemini Deep T...
Google launches AI Vulnerability Reward Program for AI system flaws
Security Tool/Service
First: 07.10.2025 16:19
Last: 07.10.2025 16:19
Sources 1
About this happening:
**Google** launched a dedicated **AI Vulnerability Reward Program** that expands its security bounty coverage to **AI system flaws** across flagship products, raising the maximum...
Google launches AI Vulnerability Reward Program for AI system flaws
Security Tool/ServiceAbout this happening: **Google** launched a dedicated **AI Vulnerability Reward Program** that expands its security bounty coverage to **AI system flaws** across flagship products, raising the maximum...
Auto Exploit LLM-assisted exploit generation research
Technical Analysis
First: 29.08.2025 16:01
Last: 29.08.2025 16:01
Sources 1
About this happening:
Researchers built **Auto Exploit**, an AI-driven system that generated proof-of-concept exploits for **14 open source vulnerabilities** in as little as **15 minutes**, compressing...
Auto Exploit LLM-assisted exploit generation research
Technical AnalysisAbout this happening: Researchers built **Auto Exploit**, an AI-driven system that generated proof-of-concept exploits for **14 open source vulnerabilities** in as little as **15 minutes**, compressing...
Timeline
-
26.08.2025 13:19 1 articles · 9mo ago
Anamorpher open source release
Initial DisclosureTrail of Bits released Anamorpher, an open source tool that lets researchers craft and visualize image scaling attacks against AI systems. The tool helps reproduce indirect prompt injection paths where a malicious prompt hidden in a high-resolution image becomes visible after preprocessing downscales it and may then be interpreted by an AI model as a legitimate instruction.
Show sources
- AI Systems Vulnerable to Prompt Injection via Image Scaling Attack — www.securityweek.com — 26.08.2025 13:19