All news with #mcp security tag

🔍 Security researchers warn major gaps in IT oversight and rising vulnerabilities are increasing risk across the AI agent ecosystem. Reco analyzed enterprise telemetry, MCP servers, and NVD disclosures in its report, The State of Agent Security 2026, finding 80% of AI tools lack governance and SMBs average 414 unsanctioned tools per 1,000 employees. The study found many MCP servers allow shell execution, file access and outbound network calls, with numerous tools exposed without authentication. Vulnerability disclosures have also surged, straining patch programs.

🛡️ Marimo fixed a high-severity code injection that let a crafted notebook supply a malicious Model Context Protocol (MCP) command executed as a local subprocess when opened in edit mode. Tracked as CVE-2026-75149 and scored ~8.7–8.8, the flaw affected versions prior to 0.23.15 and required user interaction but no attacker authentication. Marimo released version 0.23.15 and later versions to remediate the issue.

🔒 The Model Context Protocol (MCP) enables AI agents to access tools and data across enterprise systems, but the MCP server often stores credentials, tokens and keys that can expose an organization if mishandled. Common risks include plaintext config files, credential sprawl, prompt injection, over-permissioning and untrusted exposed servers. Mitigations include centralizing secrets, using short-lived credentials, enforcing least privilege, human approval for sensitive actions, end-to-end encryption, thorough logging and inventorying MCP servers.

🔒 Cloudflare announces new Cloudflare One capabilities to detect and control Model Context Protocol (MCP) traffic. These features let administrators identify which users and servers are generating MCP requests, distinguish Portal-mediated connections from direct ones, and block unauthorized direct connections on managed network paths. The update combines Gateway protocol signals with MCP Server Portals to help teams find shadow MCP servers and enforce Portal-only access to trusted MCP endpoints.

🔐 Security researchers say the real risk with AI agents lies less in the model and more in the surrounding harness — the code that turns model output into actions. Vulnerabilities in harness architecture, implementation choices, and the expanding supply chain of skills and plugins have enabled credential theft, code execution, and persistent malware. Experts urge CISOs to inventory harnesses, restrict their permissions, and independently test vendor claims to reduce exposure.

🛡️ A new technique called GhostSplice shows how a malicious Model Context Protocol (MCP) server connected to an AI coding assistant can exfiltrate SSH keys, environment secrets, source code, and customer data by splitting a theft into harmless-looking fragments. ASSET Research Group tested the approach in isolated projects using fake credentials and found that splitting the request across tool descriptions, results, or server-initiated sampling raised compliance dramatically for many models. The attack relies on developers connecting a hostile MCP server and the agent already having access to the target files.

🔒 AWS WAF now offers the Salt Security managed rule group, available via AWS Marketplace as Salt Managed Rules for AWS WAF - AI Agent & API Security. The group provides detection and mitigation for API-focused attacks and traffic from AI agents and Model Context Protocol (MCP) endpoints without requiring customers to write custom rules. It addresses threats such as credential brute force, excessive GraphQL queries, SSRF, prototype pollution, and JWT anomalies while labeling MCP traffic and sensitive request attributes. Customers can subscribe and add the rule group to a web ACL directly in the AWS WAF console; pricing and versioning are managed by Salt Security through AWS Marketplace.

🔎 Cloudflare outlines the rise of software agents as a new class of web visitors and presents the concept of an "Agentic Internet" built to be readable, discoverable, callable, and payable. The post describes open standards and tools—like Web Bot Auth, PACT, Markdown for Agents, WebMCP, and Monetization Gateway—that enable cooperation between agents and domain owners. Cloudflare positions itself as a neutral platform offering these primitives while advocating for an open, standards-based future.

🛠️ The MCP 2026-07-28 specification makes the Model Context Protocol fully stateless, removing session handshakes and Mcp-Session-Id requirements. The update simplifies deployment by allowing MCP servers to run on request-scoped infrastructure like Cloudflare Workers, while preserving elicitation via Multi Round-Trip Requests and tightening authorization with CIMD and RFC 9207 issuer checks. SDKs and migration guidance accompany the release.

🔧 Cloudflare is launching a developer preview of WebMCP, a browser standard that exposes a document.modelContext surface to enable AI agents to interact with sites without changing origin code. With a single toggle in the Cloudflare Dashboard, an edge injection adds a small bridge script that registers tool packs in the visitor's browser. Two packs—Content Credentials and Site MCP Server—run entirely client-side in this preview, while future packs may call edge workers for heavier tasks.

🛡️ Cloudflare describes WriteGuard, a centralized policy, attribution, and auditing layer for MCP servers now available in private beta. It sits between MCP clients and tool handlers to classify calls by risk, attach agent attribution, scrub audit events, and block dangerous actions before execution. The post explains how WriteGuard integrates with existing identity flows, per-tool configuration, and an asynchronous audit Worker to make agentic writes visible and controllable across MCP-connected systems.

🔒 A critical vulnerability in the open-source AI agent platform Ruflo (CVE-2026-59726, "RufRoot") allows unauthenticated attackers to exploit an exposed Model Context Protocol (MCP) bridge and gain full control of enterprise AI environments. Researchers at Noma Security showed a single HTTP request to the bridge’s /mcp endpoint can execute code, steal LLM API keys, access conversations, hijack agents, and poison persistent AI memory. Ruflo issued a rapid patch that binds the MCP bridge to loopback and enforces failure-closed behavior, while researchers urged immediate firewall and credential remediation.

🛡️ Researchers disclosed a critical vulnerability (CVE-2026-59726) in Ruflo, an open-source agent orchestration harness for Anthropic Claude Code and OpenAI Codex, that permitted unauthenticated remote code execution. The flaw, present in versions before 3.16.3, exposed an unauthenticated Model Context Protocol (MCP) bridge on port 3001 by default due to docker-compose binding to 0.0.0.0. Exploitation allowed attackers to run shell commands, steal LLM API keys, read conversations, poison AI memory, and persist backdoors. The maintainer released fixes after disclosure, changing the MCP binding to loopback, gating execution controls, and enabling MongoDB authentication.

🔍 Enterprises face rising attacks that exploit AI agents already present in their environments. These agents — coding assistants and CLI tools like Claude Code CLI, Gemini CLI, and Amazon Q CLI — can read files, run commands, and install packages, making them attractive targets when run with auto-approval. Real-world incidents, including the s1ngularity Nx npm compromise and the AgentJacking/Sentry experiments, show how prompt injection, compromised tool metadata, and unsecured MCP servers can lead to secret harvesting and covert exfiltration. Defenders must treat agents as potentially untrusted and adapt controls and monitoring accordingly.

🛡️ A hidden HTML comment in an Azure DevOps pull request can instruct a reviewer's AI coding agent to act beyond an attacker's privileges, leaking sensitive data. The flaw exists because the MCP server returns PR descriptions without the spotlighting guardrail applied elsewhere, so the agent receives hidden instructions the human reviewer cannot see. Manifold Security demonstrated a proof-of-concept that chains permitted agent calls to read cross-project resources and exfiltrate content using reviewer credentials. Microsoft acknowledged the report and recommended limiting project access and reviewing changes before running AI tools.

🛡️ Hidden text on a web page allowed Kiro, AWS's agentic coding IDE, to rewrite its mcp.json configuration and execute attacker-controlled commands on a developer's machine without a usable approval step. Researchers at Intezer and Kodem Security showed that asking Kiro to summarize or fetch a page could inject setup instructions in one-pixel white text, causing Kiro to register and launch a malicious Model Context Protocol server. AWS patched the vulnerability by protecting sensitive paths and adding platform-enforced approval checks.

🔒 This post describes a reference implementation using Cedar on AWS to prevent silent privilege escalation in multi-agent AI delegation chains. It outlines a three-layer policy model—agent-to-tool, agent-to-agent delegation, and originating user authorization—using verified token claims and HMAC-signed context. The architecture uses an MCP adapter Lambda and a Cedar evaluator Lambda to enforce policies sequentially and halt on the first deny. It includes schema, entity registrations, policy examples, deployment steps, and end-to-end test scenarios demonstrating how the model enforces least privilege.

🛡️ New Microsoft research shows attackers can hijack AI agents by poisoning a tool's description so the agent quietly exfiltrates company data. The attack leverages MCP tool descriptions—plain text that agents read—to inject hidden instructions, allowing malicious actions without obvious rule violations. Microsoft recommends treating tool descriptions as system prompts, restricting approved tools, enforcing human approval for risky actions, and monitoring agent identities and behavior.

🔗 The Gemini Enterprise Agent Platform remote MCP server lets external AI agents securely access Google Cloud Agent Platform resources. It acts as a standardized bridge so tools like Antigravity CLI or Claude Code can call models in Model Garden, manage Notebooks, and use shared prompts without leaving the IDE. Enable the API, configure your client, and use provided Toolset Endpoints to start integrating quickly while maintaining governance.

🛡️ This Microsoft Incident Response post examines an attack pattern targeting Model Context Protocol (MCP) tools, where poisoned tool metadata causes agentic AI to perform unauthorized actions. It outlines a playbook for detecting, containing, and preventing these attacks using Microsoft security controls and maps techniques to the OWASP Top 10 for Agentic Applications. The guidance emphasizes treating MCP servers as supply-chain dependencies, reviewing tool descriptions as prompts, and applying least agency controls.