Agentic AI · MCP security · near real time

Intelligence and a CISO framework for securing agentic AI

We track the latest agentic-AI and AI-cybersecurity threats worldwide and host an authoritative reference framework for securing agentic AI and Model Context Protocol (MCP) deployments — framed around identity, permissions, and blast radius.

Browse the threat feed Explore the 8-pillar framework Collaborate with us

Latest in the feed

Severity: CriticalTool-poisoning: hidden instructions in MCP tool descriptions hijack agents
Severity: HighCoding agent leaks repo secrets after reading a poisoned issue comment
Severity: HighPlaintext API keys in committed .mcp.json files found across public repos
Severity: CriticalPopular MCP server turns malicious in a point release (rug pull)
Severity: CriticalUnauthenticated RCE in MCP server bound to 0.0.0.0 over SSE

Sample data. Showing illustrative sample items as a fallback — the live feed is not available right now.

Critical now

View all

The highest-severity items requiring CISO attention.

CriticalAdvisory·MCP/tool abuse

Tool-poisoning: hidden instructions in MCP tool descriptions hijack agents

Researchers show that a malicious MCP server can embed adversarial instructions inside a tool's description metadata, which the host model reads as trusted context. Connected agents can be coerced into exfiltrating secrets or invoking other tools without the user ever seeing the injected text.

28 May 2026Global

CriticalAdvisory·Supply chain

Popular MCP server turns malicious in a point release (rug pull)

A widely installed community MCP server shipped a minor update that added covert data-collection logic to a previously benign file tool. Because most deployments had auto-approval enabled, the malicious version propagated before maintainers pulled it.

30 Apr 2026Global

CriticalVulnerability·MCP/tool abuse

Unauthenticated RCE in MCP server bound to 0.0.0.0 over SSE

A flaw in an MCP server's Server-Sent Events transport allows command execution when the service is bound to 0.0.0.0 without authentication. Internet-exposed instances were discoverable through routine scanning of /sse endpoints.

22 Apr 2026GlobalCVE-SAMPLE-001

Trend snapshot

Distribution across the current feed (12 items).

Severity: Critical: 3
Severity: High: 4
Severity: Medium: 3
Severity: Low: 1
Severity: Info: 1

The 8-pillar CISO framework

A decision-grade reference for securing agentic AI and MCP — from discovery and gateway architecture to zero-trust identity, sandboxing, and post-quantum resilience.

Read the framework