Sigma: Agentic AI MCP Semantic Tool Poisoning & Schema Override — Detection Rule
Rule Telemetry & Engineering Metadata
Section titled “Rule Telemetry & Engineering Metadata”| Evaluation Dimension | Observed Telemetry / Attribute | Epistemic Status |
|---|---|---|
| Detection Rule Identifier | DET-SIGMA-AI-AGENT-MCP-POISON | Canonical Rule Node |
| Detection Engine | SIGMA | Multi-Format Standard |
| Severity Classification | HIGH | Calibrated Alert Priority |
| Target Technology Surface | Model Context Protocol (MCP) Runtimes & Agent Ecosystems | Monitored Platform |
| Test & Validation Status | VALIDATED | Verification Assurance |
| False-Positive Risk Index | MEDIUM | Operational Noise Rating |
| Associated Weaponized CVEs | CVE-2026-101861 | Target Vulnerabilities |
| Attributed Adversary Syndicates | CLOSEDQUORUM Threat Cluster | Threat Origin |
| Linked Intrusion Campaigns | CAMP-GLOVE-2026 | Active Wave |
| Linked Exploit Chains | Agentic Tool Execution Injection to Kubernetes Cluster Takeover | Chained Topology |
| Analytical Confidence | 95% | Rule Soundness Score |
| Primary Evidence Sources | SRC-CISCO-TALOS, SRC-GITHUB-ADVISORIES | Intelligence Attestation |
Technical Detection Context
Section titled “Technical Detection Context”Monitors filesystem integrity on developer workstations and AI agent runner hosts to detect unauthorized insertion of suspicious tool executables inside MCP configuration manifests.
Detection engineering requires proactive alignment between known offensive tradecraft and defensive sensor placement. This rule intercepts adversarial attempts before lateral proliferation or host compromise can solidify.
Production Detection Signature
Section titled “Production Detection Signature”title: MCP Tool Definition Schema Modification by Untrusted Processid: c8945b12-9901-4188-b712-202609070001status: stabledescription: Detects modifications to Model Context Protocol configuration files or JSON-RPC tool declarations from non-IDE binaries.references: - https://hermes-codex.org/agentic-attack-patterns/aap-004-semantic-tool-poisoning/author: Hermes Detection Engineeringdate: 2026-09-29logsource: category: file_event product: alldetection: selection_target: TargetFilename|endswith: - '/mcp.json' - '/mcp_config.json' - '/claude_desktop_config.json' selection_content: EventData|contains: - '"tools":' - '"command": "curl"' - '"command": "nc"' - '"command": "powershell"' filter_trusted: Image|endswith: - '/code' - '/cursor' - '/windsurf' condition: selection_target and selection_content and not filter_trustedfalsepositives: - Custom local developer script provisioning environmentslevel: highSOC Triage & Prescriptive Remediation
Section titled “SOC Triage & Prescriptive Remediation”Enforce digital signature verification on MCP server definitions and lock agent capability configurations into read-only filesystems.
- Isolation Directive: Immediately isolate hosts matching this alert from the internal VLAN to prevent lateral pivoting.
- Memory & Log Preservation: Capture volatile RAM dumps and forward local daemon journals to an immutable central repository.
- Patch Verification: Validate that all emergency vendor updates associated with CVE-2026-101861 are deployed across all edge gateways.