AWS and Cisco: AI Registry scans MCP and A2A agents via YARA, LLM semantic analysis and Cisco proprietary scanners

AWS and Cisco AI Defense announced on May 13, 2026 an integration that addresses enterprise AI security gaps through AI Registry — an open control plane for unified governance of MCP servers, Agent-to-Agent agents and AI Agent Skills components. The partnership delivers three scanner technologies and a centralized audit trail for regulatory obligations.

How does AI Registry scan new MCP servers?

AI Registry automatically scans MCP servers at registration, analyzing tool descriptions and schemas before a component becomes operational. Vulnerable servers receive a security-pending tag and remain disabled until an administrator approves a review. The approach is preventive — it stops insecure agent components from being deployed, rather than reacting only after an incident.

What does the A2A Scanner specifically look for?

The A2A Scanner analyzes agent capability declarations and communication patterns through defined checklists. Detected threats include identity spoofing, prompt injection in metadata, hardcoded credentials, data exfiltration endpoints and SSRF patterns. The scanner also validates A2A specification compliance — an agent that does not follow the protocol is flagged before deployment.

What three scanner technologies run simultaneously?

The system uses three complementary approaches for full coverage:

• YARA Analyzer — pattern-based detection of SQL injection, command injection and hardcoded credentials through classic signature analysis
• LLM Semantic Scanning — AI-powered analysis via Amazon Bedrock detects sophisticated threats in tool logic and agent behavior (something signature tools cannot capture)
• Cisco Proprietary Scanners — advanced threat detection combining threat intelligence and deep code analysis

What does the audit trail ensure?

AI Registry maintains a complete security audit history for regulatory compliance requirements (SOX, GDPR), stores scan results in a central datastore and integrates with enterprise workflows — ServiceNow, Slack, SIEM systems and compliance dashboards. The goal is that every enterprise audit query can trace every agent from registration to production.

Target customers are AWS and Cisco AI Defense subscribers managing tens to hundreds of AI agents in enterprise scenarios — banks, the public sector and healthcare, where supply-chain risk from agentic architectures is becoming a regulatory concern.