LLM Security

75 posts in LLM Security · Browse all posts

- Agentic AI Security
- LLM Security
OWASP Agentic Skills Top 10 explained: the ten agent skill risks, and which to fix first
August 25, 2026
OWASP's newest Top 10 covers the skill layer: prose files agents load and act on. What all ten risks mean, and the order to fix them.
- Agentic AI Security
- LLM Security
Zero-click Grok data theft: Cryptographic Context Injection attack leaks chat histories
August 20, 2026
Cryptographic Context Injection ships attacker instructions as ciphertext the model decrypts in its own sandbox. It leaked full Grok chat histories, zero-click.
- Agentic AI Security
- LLM Security
A hole in every one: bypassing the open source AI skill scanners
July 30, 2026
We ran eight open source AI skill scanners against real attacks. A malicious skill got past all eight, including the current OASB leaderboard leader.
- Agentic AI Security
- LLM Security
The AI agent sandbox escape that breached Hugging Face: what happened, and what to fix
July 27, 2026
An OpenAI benchmark agent escaped its sandbox and breached Hugging Face, logging 17,000+ actions. What the evidence shows, what's unproven, and what to fix.
- LLM Security
We broke GPT-5.4 safety with 10 examples and 2 words using a new attack technique — IICL
April 23, 2026
IICL is a structural jailbreak bypassing GPT-5.4 safety with just two words.
- Agentic AI Security
- LLM Security
From chatbots to digital workers: Managing the business risks of agentic AI
February 18, 2026
AI agents don’t just suggest transfers — they execute them.
- Agentic AI Security
- LLM Security
Why chatbot red teaming won’t save you from agentic AI disasters
February 3, 2026
Traditional chatbot red teaming leaves 85% of the agentic AI attack surface exposed.
- Agentic AI Security
- LLM Security
Cascading failures in agentic AI: the definitive OWASP ASI08 security guide
January 4, 2026
The definitive technical guide to cascading AI failures aligned with OWASP agentic AI security guidelines: their mechanisms, manifestations, and mitigations.
- Agentic AI Security
- LLM Security
- MCP Security
Security Risks of the Model Context Protocol: Can Autonomous Agents Handle Adversarial Testing? Conversation with ChatGPT, Claude, Grok & Deepseek
August 28, 2025
Explore key findings on MCP Security from ChatGPT, Claude, Grok, and Deepseek—shared risks, conflicting views, and critical blind spots.
- LLM Security
Towards Secure AI Week 33 — Lenovo Chatbot Breach, PROMISQROUTE in GPT-5, NIST AI Security Overlays, EU AI Priorities, and Grok Privacy Leak
August 25, 2025
This week in AI security: the Lenovo chatbot breach, PROMISQROUTE in GPT-5 routers, NIST's AI security overlays, EU priorities, and a Grok privacy leak.
- LLM Security
- GenAI Security
Lenovo AI Chatbot Incident: Critical XSS Vulnerability Exposes Enterprise AI Security Gaps
August 22, 2025
A single 400-character prompt triggered a critical XSS flaw in Lenovo's AI chatbot Lena, showing how simple exploits become enterprise security incidents.
- LLM Security
PROMISQROUTE: GPT-5 AI Router Novel Vulnerability Class Exposes the Fatal Flaw in Multi-Model Architectures
August 19, 2025
PROMISQROUTE shows GPT-5 router flaw letting attackers force weaker models, bypass safety, and expose hidden risks in multi-model AI.

More research

AIRQ: AI Risk Quadrant — independent AI agent security rating

AIRQ: AI Risk Quadrant — independent AI agent security rating

AIRQ: AI Risk Quadrant — independent AI agent security rating

LLM Security

OWASP Agentic Skills Top 10 explained: the ten agent skill risks, and which to fix first

Zero-click Grok data theft: Cryptographic Context Injection attack leaks chat histories

A hole in every one: bypassing the open source AI skill scanners

The AI agent sandbox escape that breached Hugging Face: what happened, and what to fix

We broke GPT-5.4 safety with 10 examples and 2 words using a new attack technique — IICL

From chatbots to digital workers: Managing the business risks of agentic AI

Why chatbot red teaming won’t save you from agentic AI disasters

Cascading failures in agentic AI: the definitive OWASP ASI08 security guide

Security Risks of the Model Context Protocol: Can Autonomous Agents Handle Adversarial Testing? Conversation with ChatGPT, Claude, Grok & Deepseek

Towards Secure AI Week 33 — Lenovo Chatbot Breach, PROMISQROUTE in GPT-5, NIST AI Security Overlays, EU AI Priorities, and Grok Privacy Leak

Lenovo AI Chatbot Incident: Critical XSS Vulnerability Exposes Enterprise AI Security Gaps

PROMISQROUTE: GPT-5 AI Router Novel Vulnerability Class Exposes the Fatal Flaw in Multi-Model Architectures

Blog categories

Explore Our Blog

LLM Security

LLM Security posts

OWASP Agentic Skills Top 10 explained: the ten agent skill risks, and which to fix first

Zero-click Grok data theft: Cryptographic Context Injection attack leaks chat histories

A hole in every one: bypassing the open source AI skill scanners

The AI agent sandbox escape that breached Hugging Face: what happened, and what to fix

We broke GPT-5.4 safety with 10 examples and 2 words using a new attack technique — IICL

From chatbots to digital workers: Managing the business risks of agentic AI

Why chatbot red teaming won’t save you from agentic AI disasters

Cascading failures in agentic AI: the definitive OWASP ASI08 security guide

Security Risks of the Model Context Protocol: Can Autonomous Agents Handle Adversarial Testing? Conversation with ChatGPT, Claude, Grok & Deepseek

Towards Secure AI Week 33 — Lenovo Chatbot Breach, PROMISQROUTE in GPT-5, NIST AI Security Overlays, EU AI Priorities, and Grok Privacy Leak

Lenovo AI Chatbot Incident: Critical XSS Vulnerability Exposes Enterprise AI Security Gaps

PROMISQROUTE: GPT-5 AI Router Novel Vulnerability Class Exposes the Fatal Flaw in Multi-Model Architectures

Blog categories

Explore Our Blog