PPhying中文
Curated security research, vulnerabilities, advisories and tools for practitioners.

How MCP servers can steal your conversation history

blog.trailofbits.com | blog | #mcp | #prompt-injection | #llm-security | #ai-agents | #data-exfiltration | #trail-of-bits
📄 Raw .md⬇️ Download .md🔗 Original ↗

Summary

Malicious MCP servers can inject trigger phrases into tool descriptions to exfiltrate entire conversation histories and steal sensitive credentials and IP.
Published
Collected

Related coverage

blog · blog.trailofbits.com
State divergence enables unauthorized access
This coverage gives security teams current context for monitoring, validation, and remediation.
blog · about.gitlab.com
New MCP tools help platform teams scale automation safely
GitLab 19.4 expands the GitLab MCP server tool set so agents can run pipelines, open merge requests, and triage vulnerabilities, while writes, merges, and deletes still wait for a human reviewer.
blog · wiz.io
Growing the WIN AI Ecosystem with Agent Integrations
Wiz MCP agent integrations now let partner AI agents pull live Wiz context—Issues, findings, attack paths—on demand from customer tenants, with the WIN partners endpoint GA for building integrations.
blog · wiz.io
Building an AI Detection Engine That Understands Agent Intent
Wiz outlines a detection pipeline that analyzes AI agent input and output telemetry to trace shifts in intent, using the OpenAI-disclosed Hugging Face breach to show what legacy tools miss.
blog · wiz.io
Build AI Security Agents with Wiz MCP
Wiz MCP is now GA, letting AI assistants and custom agents query the Wiz Security Graph, reuse Red, Blue and Green agent analyses, and run AI Skills workflows for triage and remediation.
blog · netspi.com
How App Integration Transactions Increase the Attack Surface of LLMs
Learn how OpenAI's AppsSDK, AgentKit, and "Buy It" turn LLMs into transactional agents—expanding security risks from rapid rollout, prompt injection, and access control gaps.
Back