trust-boundary

Build your own security agents. Open-source framework for agents with live, read-only access to your infrastructure, with no path to widen it. Reasons across AWS, GCP, Azure, Kubernetes, GitHub and GitLab as one system.

An open-source engineering governance standard defining trust boundaries for conversational AI agents in high-stakes domains. MIT Licensed.

Noisegate: a differential privacy gateway that lets an untrusted LLM agent query sensitive data over MCP (Model Context Protocol), with a formal guarantee no individual's record can leak even if the agent is adversarial - enforcement lives in trusted code below the model, validated by a runnable attack gallery.

Turn untrusted source material into compact, attributed AI skills through evidence gates and deterministic safety checks.

Omega Walls — a deterministic runtime trust boundary for RAG and AI agents that models cumulative prompt-injection, secret-exfiltration, and tool-abuse pressure before untrusted content reaches context or tools.

detect structural trust boundary vulnerabilities that enable cascading supply chain compromise

A systems-security framework for evaluating trust in open-weight LLM deployments — air-gapped environments, hidden backdoors, and software supply chain integrity.

Multi-interface trust boundary protocol for AI agent systems

Revenue triage CLI + public fit-check issue for products with attention but no paid conversion.

Protect applications by managing untrusted content with a stateful runtime that enforces risk states and controls actions in RAG and agent systems.

Why every AI agent is vulnerable to prompt injection — and a proposed four-layer security architecture to fix it. HMAC auth + intent-based permissions + sandboxed execution + result audit.

A behavioral trust boundary for API keys and MCP tools. Concept and reference architecture, published in 12 languages.

Kinetic Gain, Trust Boundary Declaration spec (JSON Schema 2020-12). Read-only by default; writes are named, audited grants; live requires a fetchable attestation.

Typed validated values, explicit trust transitions, exhaustive confidence levels, and per-value TTL freshness for OCaml.

Policy-enforcing trust boundary that makes slashable signing impossible for PoS validators.

Typed validated values, explicit trust transitions, exhaustive confidence levels, and per-value TTL freshness for Haskell.

Self-evolving prompt-injection defense: breach -> synthesize antibody -> two-sided gate -> promote -> broadcast.

Typed validated values, explicit trust transitions, exhaustive confidence levels, and per-value TTL freshness for Swift.

Loss-aware receiver-side admission kernel for AI agents: explicit UNKNOWN, exact evidence challenges, least-authority scope, and signed DSSE/Ed25519 receipts. Synthetic offline prototype.

Typed validated values, explicit trust transitions, exhaustive confidence levels, and per-value TTL freshness for F#.