llm
118 articles with this tag
MEDIUM
INFO
INFO
MEDIUM
MEDIUM
MEDIUM
HIGH
HIGH
INFO
INFO
MEDIUM
HIGH
CRITICAL
INFO
MEDIUM
INFO
INFO
CRITICAL
HIGH
HIGH
INFO
CRITICAL
CRITICAL
CRITICAL
HIGH
INFO
MEDIUM
HIGH
INFO
INFO
INFO
INFO
CRITICAL
CRITICAL
HIGH
HIGH
MEDIUM
HIGH
CRITICAL
CRITICAL
LOW
MEDIUM
MEDIUM
MEDIUM
INFO
INFO
INFO
HIGH
CRITICAL
HIGH
CRITICAL
LOW
MEDIUM
INFO
INFO
HIGH
CRITICAL
CRITICAL
INFO
INFO
CRITICAL
MEDIUM
HIGH
MEDIUM
CRITICAL
CRITICAL
HIGH
INFO
HIGH
INFO
CRITICAL
HIGH
INFO
MEDIUM
INFO
HIGH
MEDIUM
INFO
HIGH
MEDIUM
HIGH
HIGH
MEDIUM
INFO
LOW
MEDIUM
INFO
MEDIUM
LOW
INFO
MEDIUM
INFO
INFO
LOW
INFO
INFO
MEDIUM
INFO
INFO
INFO
PuzzleMask: Abusing Plain Prose as a Covert AI Attack Vector
Perturbation Probing: A New Diagnostic for the Fragility of LLM Safety
Stealing AI Reasoning Traces
Chatbot leaked a planted phone number and failed 58% of prompt injection attempts - jailbreak and extraction checks stayed clean
[NEU] [mittel] LiteLLM: Schwachstelle ermöglicht Umgehen von Sicherheitsvorkehrungen
[NEU] [mittel] vllm: Schwachstelle ermöglicht Denial of Service
[NEU] [hoch] Langflow OSS: Mehrere Schwachstellen
Grok exfiltrates user data when malicious instructions are encrypted
LLMs and Contextual Integrity
AI-Assisted Network Threat Hunting: See It, Enrich It, Ask It-BHIS Webcast
AI struggles to patch vulns without adult supervision
Prompt injection remains top LLM threat, OWASP report finds
AI Recommendation Poisoning: How "Ask AI" Buttons Silently Alter LLM Memory
Bypassing AI guardrails is so easy a script kiddie can do it
[NEU] [mittel] Flowise: Mehrere Schwachstellen
Claude Opus 5 sharpens coding and cybersecurity work on AWS
How to Secure AI Applications in Production
How an OpenAI benchmark test turned into a real-world cyberattack
[NEU] [mittel] Ollama: Schwachstelle ermöglicht Denial of Service
OpenAI Says Its AI Models Broke Loose and Hacked Hugging Face
Using LLMs to Find and Prioritize Vulnerabilities Is No Easy Task
This AI agent autonomously hacked a network, adapted on the fly, and demanded a ransom
Smooth AI criminal drives 'first' end-to-end agentic ransomware attack
Somebody told DeepSeek to build in-browser ransomware and it gleefully complied
[NEU] [hoch] LiteLLM: Mehrere Schwachstellen
Snyk VulnBench JS 1.0: Can LLMs Find the Same Bugs Twice?
[NEU] [mittel] Flowise: Schwachstelle ermöglicht Umgehen von Sicherheitsvorkehrungen
[NEU] [mittel] Ollama (Quantization Engine): Schwachstelle ermöglicht Offenlegung von Informationen
Measuring LLM system prompt extraction (OWASP LLM07) against ground truth, across 4 models
Interesting Paper Exploring Prompt Injection
LLM security advice looks solid until you check the hard cases
Assessing Automated Prompt Injection Attacks in Agentic Environments
[NEU] [hoch] vllm: Schwachstelle ermöglicht Umgehen von Sicherheitsvorkehrungen
[NEU] [hoch] Langflow: Schwachstelle ermöglicht Denial of Service und Offenlegung von Informationen
[NEU] [mittel] vllm: Mehrere Schwachstellen
AI-driven computer worm demonstrates autonomous network exploitation
Autonomous AI-driven worm can reason its way through corporate networks
Interesting- What LLM vuln research looks like
Exploit Code Published for Critical Flowise RCE Vulnerability
Attackers Use LLM Agent for Post-Exploitation After Marimo CVE-2026-39987 Exploit
Anthropic launches Claude Opus 4.8, prepares Mythos-class models for all customers
I evaluated 5 LLM agents on patching real-world CVEs. Here is what I found.
All Major LLMs Exposed to Multi-Turn Manipulation, Warn Researchers
RCE in Strix Agent(Sandbox): A practical guide to prompt injections with impact
Prompt Injection finally broke my brain a little. My first article as a security student.
Why Policy in Amazon Bedrock AgentCore chose Cedar for securing agentic workflows
Sandyaa: Open-source autonomous security bug hunter
[NEU] [hoch] LangChain: Schwachstelle ermöglicht Manipulation von Dateien und Offenlegung von Informationen
Ollama Out-of-Bounds Read Vulnerability Allows Remote Process Memory Leak
[NEU] [hoch] LiteLLM: Schwachstelle ermöglicht Codeausführung und Offenlegung von Informationen
[NEU] [hoch] LiteLLM: Schwachstelle ermöglicht Umgehen von Sicherheitsvorkehrungen
The Bot Left a Fingerprint: Detecting and Attributing LLM-Generated Passwords
Indirect prompt injection is taking hold in the wild
Researchers Simulated a Delusional User to Test Chatbot Safety
Groundcover expands AI Observability for LLM interactions
VU#518910: Ollama GGUF Quantization Remote Memory Leak
SGLang CVE-2026-5760 (CVSS 9.8) Enables RCE via Malicious GGUF Model Files
VU#915947: SGLang is vulnerable to remote code execution when rendering chat templates from a model file
Unweight: how we compressed an LLM 22% without sacrificing quality
Building the foundation for running extra-large language models
[NEU] [hoch] Flowise: Mehrere Schwachstellen
Command integrity breaks in the LLM routing layer
Unpatched RAGFlow Vulnerability Allows Post-Auth RCE
[NEU] [mittel] vllm: Mehrere Schwachstellen
Max severity Flowise RCE vulnerability now exploited in attacks
Critical Flowise Vulnerability in Attacker Crosshairs
DeepZero: An automated LLM/Ghidra pipeline for finding BYOVD zero-days in Windows drivers
Google Workspace’s continuous approach to mitigating indirect prompt injections
Vim and GNU Emacs: Claude Code helpfully found zero-day exploits for both
An attack class that passes every current LLM filter
LangChain, LangGraph Flaws Expose Files, Secrets, Databases in Widely Used AI Frameworks
A nearly undetectable LLM attack needs only a handful of poisoned samples
Training an AI agent to attack LLM applications like a real adversary
Novee introduces autonomous AI red teaming to hunt LLM vulnerabilities
Microsoft details AI prompt abuse techniques targeting AI assistants
VU#665416: SGLang (sglang) is vulnerable to code execution attacks via unsafe pickle deserialization
Researchers Discover Major Security Gaps in LLM Guardrails
Auditing the Gatekeepers: Fuzzing "AI Judges" to Bypass Security Controls
Claude Used to Hack Mexican Government
Red Teaming LLM Web Apps with Promptfoo: Writing a Custom Provider for Real-World Pentesting
[UPDATE] [hoch] vllm: Mehrere Schwachstellen
[UPDATE] [hoch] vllm und PyTorch: Schwachstelle ermöglicht DoS und potenzielle Codeausführung
Reverse CAPTCHA: Evaluating LLM Susceptibility to Invisible Unicode Instruction Injection
Reverse CAPTCHA: Evaluating LLM Susceptibility to Invisible Unicode Instruction Injection
I made Mistral believe Donald Trump runs OpenAI, here's how
Arkanix Stealer pops up as short-lived AI info-stealer experiment
Anthropic rolls out embedded security scanning for Claude
Hacking LLMs: How to Prevent Prompt Injection Attacks Now
LLMs change their answers based on who’s asking
Virtana expands MCP Server to bring full-stack enterprise context to AI agents
Side-Channel Attacks Against LLMs
Prompt Injection Standardization: Text Techniques vs Intent
Security audit for LLM skill files: skillaudit.sh
The Promptware Kill Chain
I built a free, open-source platform to learn GenAI security, learning content + hands-on labs against real LLMs (beta, looking for feedback)
Lumma Stealer: Breaking down the delivery techniques and capabilities of a prolific infostealer | Microsoft Security Blog
VoidLink Linux C2 Uses LLM-Generated Malware with Kernel-Level Stealth
The OpenClaw experiment is a warning shot for enterprise AI security
The Future of AI Agent Security Is Guardrails
1Password open sources a benchmark to stop AI agents from leaking credentials