RelayShield releases smolagents tools and MITRE ATT&CK dataset for agentic security

RelayShield has released four `smolagents.Tool` classes designed to help agents evaluate their own attack surface before execution. These tools check if an MCP server is known-malicious, detect prompt-injection patterns in incoming content, scan the agent's tech stack for known CVEs, and perform bulk identity-risk scoring for domains and emails.

• The toolkit includes a Hugging Face Space for the agentic attack surface interface.
• A corresponding dataset maps MITRE ATT&CK groups and techniques.
• The system is backed by a live threat-intel pipeline monitoring 3M+ IOCs, 4,500+ malware families, and 40+ criminal marketplaces.

The release aims to provide dynamic security checks for autonomous agents rather than relying on static demonstrations.