Research & Writing

The case for public cryptographic trust infrastructure and multi-agent adversarial debate for the MCP ecosystem. Covers the SmartLoader supply chain attack, the agent identity crisis, and the design philosophy behind Credence and ThinkTank. 39 references.

How MCP standardizes the nervous system of agentic AI — and why that makes credential theft a systemic risk.

Supply chain attacks on AI systems that degrade trust silently rather than breaking things loudly.

A guide to AI data poisoning for the security-savvy.

What AI agent hijacking reveals about the future of offensive and defensive AI security.

Source code, attestation schema, scan pipeline, and MCP server implementation.

The Credence MCP server and CLI. Install it to check attestations from Claude Desktop, Claude Code, or any MCP client.

The live registry file. Trust scores, verdicts, provenance data, and Ed25519 signatures for every scanned server.

The CI workflow that runs the full Credence pipeline: collect, scan, deliberate, attest, sign, publish.

The attack that demonstrated why social signals (stars, forks, README quality) are insufficient for MCP trust decisions.

Automated security assessment for open source projects. Credence's risk tier weights are derived from Scorecard's methodology.

Keyless code signing infrastructure. Informed Credence's approach to cryptographic attestation and transparency logs.

The protocol specification that defines MCP server behavior, tool definitions, and the transport layer Credence audits.