Macaw adds a cryptographic trust layer to AI agent workflows by providing drop‑in secure SDKs that sign and authenticate every invocation, enforce policy rules, and generate tamper‑evident audit logs.
Funding
Funding not disclosed
Founders
Product
Problem
AI agents can invoke other agents, tools, and LLMs without verifiable identities or enforced permissions, creating blind spots where agents may act outside intended boundaries, spoof identities, inject malicious prompts, or tamper with logs. Existing IAM and reactive defenses cannot keep pace with the rapidly expanding attack surface of autonomous AI workflows.
Solution
Macaw provides a cryptographic trust layer that replaces standard AI client libraries with drop‑in secure adapters (e.g., SecureOpenAI, SecureAnthropic). Each agent, tool, and model is issued a signed keypair, enabling authenticated invocations, prompts, and context. The platform enforces policies defined in plain English, maps identity attributes from existing IdPs (Okta, Azure AD, Keycloak, etc.), and records tamper‑evident attestations for every operation. A real‑time console visualizes the full AI mesh, showing topology, policy details, and audit trails, allowing operators to monitor and control autonomous workflows without changing application code.
Target Audience
Primary customers are enterprise developers and security teams deploying autonomous AI agents in production environments, including financial services, healthcare, and large technology firms that require verifiable AI behavior and compliance.
Features
- Drop‑in SDKs that add cryptographic signing, policy enforcement, and audit logging to existing OpenAI, Anthropic, and LangChain integrations
- Distributed trust mesh where every participant (agent, tool, LLM) registers with a signed identity and can verify inbound requests independently
- Policy Assistant that translates natural‑language rules into enforceable MAPL policies and explains policy decisions
- Real‑time activity graph displaying agents, tools, and connections with click‑through details on permissions and provenance
- Authenticated prompts and context that protect against prompt injection and ensure permissions only narrow through workflow steps
- Integration with enterprise IdPs (Okta, Azure AD, Keycloak, Auth0) for automatic claim mapping and multi‑tenant policy scoping
- Tamper‑evident audit logs and cryptographic attestations for full traceability from prompt to response