MCP tool results need a materialization boundary
MCP tool results can include resource links, audio, images, and structured payloads. A materialization boundary keeps rich tool output from poisoning agent sessions.
Tag archive
MCP
22 articles connected to this topic.
Semantic Kernel RCE shows why AI agent tool inputs need real boundaries
Microsoft's Semantic Kernel RCE research shows how prompt injection becomes code execution when agents can influence tool parameters. Here's how to reduce the blast radius.
OpenClaw skill security checklist: what to review before you install
A practical checklist for reviewing OpenClaw skills and MCP servers before they get access to your files, accounts, shell, or APIs.
Cisco Launches DefenseClaw: Open-Source Security for the Agentic Workforce
At RSAC 2026, Cisco introduced DefenseClaw — an open-source framework that scans agent skills, verifies MCP servers, and inventories AI assets. Combined with Zero Trust for agents and NVIDIA OpenShell, it's the most comprehensive agent security stack yet.
Astrix Security Builds a Control Plane for Shadow AI Agents at RSAC 2026
Astrix Security unveils a four-method AI agent discovery engine and real-time Agent Control Plane at RSAC 2026 — combining NHI fingerprinting, EDR telemetry, and platform integrations to find and govern every shadow AI agent in the enterprise.
Snyk Launches Agent Security: Governing Autonomous Coding Agents from Dev to Production
Snyk unveils Agent Security and Evo AI-SPM GA at RSAC 2026 — a full-lifecycle enforcement architecture that secures AI coding agents like Claude Code, Cursor, and Devin across environment, artifact, and behavior, with Agent Scan, Studio, and Agent Guard.
SentinelOne Ships AI Agent Security, Red Teaming, and Auto Investigation GA at RSAC 2026
SentinelOne goes GA on four products at RSAC 2026: Prompt AI Agent Security for real-time agent governance with MCP monitoring, Prompt AI Red Teaming for continuous AI application testing, Purple AI Auto Investigation for one-click agentic SOC, and AI data pipelines that cut SIEM noise by 80%.
SentinelOne Ships Real-Time AI Agent Governance and One-Click Agentic Investigations at RSAC 2026
SentinelOne launches Prompt AI Agent Security for real-time MCP server governance, Prompt AI Red Teaming for AI app hardening, and general availability of Purple AI Auto Investigation — agentic forensic investigations that compress hours into minutes.
Cisco Extends Zero Trust to AI Agents, Open-Sources DefenseClaw at RSAC 2026
Cisco's RSAC 2026 keynote introduces Zero Trust Access for AI agents via MCP proxy, DefenseClaw open-source secure agent framework, AI Defense Explorer Edition for self-service red teaming, and six specialized SOC agents. 85% of enterprises pilot AI agents — only 5% reach production.
Databricks Ships 35 New Agentic AI Security Risks in DASF v3.0
The Databricks AI Security Framework v3.0 adds 35 agentic-specific risks and 6 new controls covering agent reasoning, memory poisoning, MCP server/client threats, and multi-agent system attacks — bringing the total to 97 risks and 73 controls.
Red Hat Shows How to Operationalize OpenClaw for the Enterprise — Without Rewriting Your Agent
Red Hat AI's 'Bring Your Own Agent' blueprint uses OpenClaw as its reference agent, adding SPIFFE identity, MCP Gateway authorization, Kata Containers isolation, and MLflow tracing — all without touching agent code.
OWASP Expands AI Security Frameworks Ahead of RSAC 2026 — With an Agentic Hackathon
The OWASP GenAI Security Project releases its most comprehensive update yet: agentic red teaming taxonomy, MCP server security guide, GenAI data security risks — plus a live agentic AI Capture the Flag at RSAC 2026.
Salt Security Maps the Agentic Attack Surface: LLMs, MCP Servers, and APIs in One Graph
Salt Security launches the industry's first platform to secure the entire agentic AI stack — mapping how LLMs reason, MCP servers connect, and APIs execute — with the Agentic Security Graph providing real-time visibility into what your AI agents can actually do.
ConductorOne Wants to Make the Governed Path to AI Faster Than the Ungoverned One
ConductorOne launches AI Access Management — a unified control plane for AI tools, agents, and MCP connections with 3,000+ hosted MCP servers, credential vaulting, and fine-grained tool call authorization. The pitch: if getting approved AI access takes 60 seconds, nobody needs shadow AI.
Two Companies Just Solved Different Halves of the AI Agent Privilege Problem
Entro Security launches Agentic Governance & Administration (AGA) for shadow AI discovery and MCP enforcement. Apono launches Agent Privilege Guard with Intent-Based Access Controls and zero standing privileges. Together, they map the full agent governance stack — both headed to RSAC 2026.
Axiory Builds a Trading Platform Where AI Agents Are the Traders
Axiory launches infrastructure for AI agents to autonomously trade FX, stocks, and ETFs via MCP. The shift from dashboards to agent-native finance has begun.
Mimecast Rewires Enterprise Security for the AI Agent Era
Mimecast's March 2026 platform overhaul introduces adaptive security policies, an AI investigation agent, and an MCP gateway — treating the human layer as the new security control plane as AI agents flood enterprise environments.
Singulr AI's Agent Pulse: Runtime Governance for Autonomous Agents and MCP Servers
Singulr AI launches Agent Pulse — a governance platform that discovers, risk-scores, and enforces policies on autonomous AI agents and MCP servers in real time. Because deploying agents without runtime controls is flying blind.
One-Third of MCP Servers Are Vulnerable to SSRF — And Hundreds Have Zero Auth
BlueRock Security scanned 7,000+ MCP servers and found 36.7% vulnerable to SSRF. Trend Micro found 492 exposed with no authentication. The MCP ecosystem has a systemic security problem.
Google Just Released a Workspace CLI Built for AI Agents — And It Has an MCP Server
Google's new open-source gws CLI gives AI agents structured access to Gmail, Drive, Calendar, Sheets, and every Workspace API. It includes 100+ agent skills and an MCP server mode that works with OpenClaw out of the box.
MCP Atlassian RCE: When Your Agent's Tools Become Attack Vectors
CVE-2026-27825 allows unauthenticated remote code execution through mcp-atlassian's Confluence integration. Another reminder that MCP tool security is the weakest link in agent deployments.
Claude Code MCP Vulnerabilities: How Cloning a Repo Could Steal Your API Keys
Two critical CVEs in Anthropic's Claude Code exploited MCP configuration to achieve remote code execution and API key theft. What OpenClaw users should know about supply chain attacks on AI agents.