Agent security bulletin
Latest News
Agent Security Weekly and the 10 latest links per topic. New marks links indexed in the last hour.
Agent Security Weekly
Stolen keys put agents at risk
What stolen vendor keys, model supply chains, and weak inputs mean for people running AI agents.
Vulnerabilities & CVEs
CVE-2026-133505: When message history is serialized and loaded again, a tool result that contains a JSON object in the serialized shape of one of Pydantic AI's file types is rebuilt as that file. A th...
Automatically indexed from the source. Read the original for details.
SoK: Assessing the Evolution of LLM Capability in Software Vulnerability Research
Automatically indexed from the source. Read the original for details.
CVE-2026-18883: ServiceNow has addressed an indirect prompt injection security issue that was identified in the Email Response AI Agent of Now Assist for CSM. We are not currently aware of malicious e...
Automatically indexed from the source. Read the original for details.
CVE-2026-125288: ServiceNow has addressed a code injection security issue that was identified in the ServiceNow AI Platform. This issue could enable an authenticated user with template editing privile...
Automatically indexed from the source. Read the original for details.
Navigating Vulnerability Hysteria in the AI Era
Automatically indexed from the source. Read the original for details.
CVE-2026-77359: Jupyter MCP Server is a Model Context Protocol server for Jupyter. Prior to 1.0.3, standalone streamable-http mode validates Host and Origin values for /mcp but does not apply the same...
Automatically indexed from the source. Read the original for details.
CVE-2026-77318: Jupyter MCP Server is a Model Context Protocol server for Jupyter. Prior to 1.0.3, standalone streamable-http mode can enforce MCP TOKEN on /mcp while leaving the state-changing /api/c...
Automatically indexed from the source. Read the original for details.
CVE-2026-133161: Summary ui With strip ansi=True , Pydantic AI Harness's ToolOutputLimits capability removes ANSI escape sequences from tool output with a pattern that takes time quadratic in the leng...
Automatically indexed from the source. Read the original for details.
CVE-2026-77283: obsidian-sync-mcp is an MCP server that gives AI agents access to Obsidian vaults. Prior to 0.5.7, deployments running without MCP AUTH TOKEN install no serverOptions.authenticate chec...
Automatically indexed from the source. Read the original for details.
CVE-2026-75612: yt-mcp is a YouTrack MCP server for Claude Code. Prior to 1.18.1, add attachment in src/yt mcp/tools/attachments.py accepts a caller-controlled file path, checks only that the file exi...
Automatically indexed from the source. Read the original for details.
Attacks & incidents
ThreatsDay: GPT-6.2 Jailbreak Flaws, BubbleLoader PNG Malware, $500M Ransom Demands + 17 More Stories
Automatically indexed from the source. Read the original for details.
AI incidents of 2026: Rogue agents, stolen keys, and other familiar stories
Automatically indexed from the source. Read the original for details.
CISA Warns of 4 Exploited Flaws Linked to Ransomware, Cryptomining, and LLMjacking Attacks
Automatically indexed from the source. Read the original for details.
Researchers take over a popular model account on Hugging Face
The Kokoro incident shows why teams should pin model downloads to a version they have reviewed.
'TheMailer' Phishing Kit Puts AI in the Driver's Seat
Automatically indexed from the source. Read the original for details.
Cursor Traces Malicious Claude Code Extensions to Third-Party Breach
Automatically indexed from the source. Read the original for details.
Ransomware Attacks Increase Behind AI, but Payments Decline
Automatically indexed from the source. Read the original for details.
Cursor Says Stolen Fireworks AI Key Led to Trojanized Claude Code Extension Incident
Automatically indexed from the source. Read the original for details.
186 Fake AI Domains Redirect Users to ClickFix Lures, Browser Malware, and Search Hijackers
Automatically indexed from the source. Read the original for details.
Inside an AI Attack Against a US Retailer
Automatically indexed from the source. Read the original for details.
Defenses & guidance
AI Pentesting Checklist for Web Apps: 15 Questions to Ask Before Choosing a Vendor
Automatically indexed from the source. Read the original for details.
External sandboxes for coding agents
Why running coding agents away from your main machine can limit what they can reach.
Webinar: See How Identity Blind Spots Put Your SOC at Risk and How to Close Them
Automatically indexed from the source. Read the original for details.
Research papers
Replication of Agentic Security Events: Audit and Investigation (ROGUE-AI)
Automatically indexed from the source. Read the original for details.
SEOYUN: Refusal-Inducing Exam Documents for AI-Glasses Pipelines
Automatically indexed from the source. Read the original for details.
Jailbreak Detection on Large Language Models via Contrastive Prototype Structures
Automatically indexed from the source. Read the original for details.
PIPETTE: Prompt Injection Purple-Teaming for Robust LLM Agents
Automatically indexed from the source. Read the original for details.
AutoPI: Taxonomy-Guided Adaptive Red Teaming for Prompt Injection in LLM Agents
Automatically indexed from the source. Read the original for details.
What Survives the Shift from Text to Pixels? Conditional Protection against Prompt Injection in Image-Rendered Documents
Automatically indexed from the source. Read the original for details.
Network-Exposed Configuration Attacks on LLM-Controlled Social Robots
Automatically indexed from the source. Read the original for details.
Securing and Auditing Long-Horizon LLM Agents: State-Transfer Auditing and Controls in the Agent Harness
Automatically indexed from the source. Read the original for details.
Dressed in Authority: Practical Black-Box Jailbreaking of MLLMs via Visual Trust Bias
Automatically indexed from the source. Read the original for details.
GapLeak: When Privacy-Preserving LLM-based Web Agents Become a Privacy Risk
Automatically indexed from the source. Read the original for details.
Other security news
Token fraud is here and it’s BAD (YouTube)
Automatically indexed from the source. Read the original for details.
Why the Harness Matters: Early Lessons From the Evo Code Scanning Agent Preview
Automatically indexed from the source. Read the original for details.
30 Years of Gatekeeper Asymmetry Doom: The Paradox that Makes Powerful LLMs More Vulnerable
Automatically indexed from the source. Read the original for details.
Java4Shell Flaw Threatens Global Enterprises, Java Application Ecosystem
Automatically indexed from the source. Read the original for details.
Quoting Daniel Stenberg
Automatically indexed from the source. Read the original for details.
Quality Isn’t The Main Worry About AI Pentesting - Data Privacy and Production Safety Are
Automatically indexed from the source. Read the original for details.
Google Finds Hackers Using AI Agents to Build and Run Exploits From Software Patches
Automatically indexed from the source. Read the original for details.
⚡ Weekly Recap: N-able 0-Days, Windows 0-Days, Chrome 0-Days, Fridge 0-Days, and More Bad Days
Automatically indexed from the source. Read the original for details.
Amazon Uncovers AI Agent Campaign Targeting LiteLLM, Dify, and Other Apps for API Keys
Automatically indexed from the source. Read the original for details.
datasette-llm 0.1a11
Automatically indexed from the source. Read the original for details.