Identity & Governance for AI Agents
AI Identity gives every autonomous AI agent its own cryptographic identity, scoped permissions, and a tamper-evident audit trail — so security, platform, and compliance teams can deploy agents with accountability built in.
AI Identity is part of the Google for Startups Cloud Program and MongoDB for Startups.
Founded
2026
Headquarters
Boulder, CO
Stage
Design Partner
Category
Agent IAM + Audit
What AI Identity Builds
AI Identity is a business-to-business software platform. We ship a hosted control plane and an open-source SDK that give organizations three things their current stack does not provide for AI agents:
Per-agent API keys
Unique, scoped aid_sk_ credentials per agent. SHA-256 hashed at rest. Zero-downtime rotation. Revoke one agent without disturbing the others.
Context-aware gateway
A deny-by-default proxy in front of your LLM providers and internal APIs. Enforces least-privilege, rate limits, spending caps, and human-in-the-loop approvals for high-risk actions.
Forensic audit trail
HMAC-SHA256 hash-chained logs, DSSE + ECDSA P-256 signed session attestations, and an offline verification CLI. Auditors verify independently of our servers.
Problems We Solve
Enterprises are deploying AI agents into production with infrastructure that was designed for humans and long-lived services — not for thousands of short-lived, autonomous workers making decisions on their behalf. That gap shows up in four places.
Shared API keys hide who did what
Teams ship one OpenAI/Anthropic key across every agent. When something goes wrong, there is no way to attribute an action to a specific agent, tool-call, or user session.
No least-privilege at the agent layer
Existing IAM and secret managers scope access per human or service account. There is no primitive for per-agent permissions, spending caps, tool allowlists, or deny-by-default enforcement.
Audit logs aren't forensic evidence
Provider-side logs are fine for debugging, but they are vendor-controlled, mutable, and incomplete. Regulators and auditors need tamper-evident, independently verifiable records.
Compliance frameworks don't map cleanly
EU AI Act Article 12, NIST AI RMF Measure, and SOC 2 CC7 all require traceability for automated decisions. Most teams have no control mapping between their runtime and the frameworks they're audited against.
Who We Serve
Our target customer is an organization deploying autonomous AI agents into production — typically through LangChain, LlamaIndex, CrewAI, AutoGen, OpenAI Agents SDK, or a custom framework — and accountable for what those agents do.
Platform & AI engineering teams
Own the agent runtime. Need per-agent keys, quota control, and a drop-in gateway that doesn't require rewriting every agent.
Security & IAM teams
Own authorization. Need deny-by-default enforcement, scoped credentials, rotation, and anomaly detection for non-human identities.
Risk, compliance & audit
Own the evidence. Need tamper-evident logs, signed attestations, and control mapping to EU AI Act, SOC 2, NIST AI RMF, GDPR, and HIPAA.
Industries we focus on first
Team
AI Identity is founder-led. Jeff drives product, engineering, and customer relationships directly — which means every design partner gets direct access to the person who built the platform.
Jeff Leva
Founder & CEO
Boulder, Colorado
Jeff has spent his career building and operating production systems in environments where failure isn't an option — cloud banking infrastructure handling $50B+ in client assets, enterprise platforms that teams run their businesses on, and the security and compliance tooling those environments demand.
AI Identity came out of a pattern he kept seeing: organizations spinning up AI agents with shared credentials, no identity layer, and no way to answer who did what when something went wrong. Jeff founded AI Identity to give those teams the identity, policy, and evidence primitives they already expect for humans — purpose-built for agents.
Team & Partners
We're selectively building our founding team and advisory network. If you've scaled a developer platform, security product, or compliance tooling and want to be close to what's being built at the AI identity layer — get in touch.
We're also in conversation with potential co-founders with deep go-to-market or enterprise security backgrounds. If that's you, let's talk.
Current Stage
AI Identity is in early launch. Core product is live and running in production on our own infrastructure. We're selectively onboarding design partners and refining the product against their real workloads before broader GA.
Live today
- Hosted control plane + dashboard
- Gateway with OpenAI, Anthropic, Gemini
- Python SDK + REST API + CLI
- HMAC hash-chained audit log with org scoping + correlation IDs
- DSSE + ECDSA P-256 signed attestations (KMS + public JWKS)
- Offline attestation verification CLI
- Human-in-the-loop approvals (Enterprise tier)
- Shadow-agent detection with Register / Block / Dismiss flows
- ABAC on agent metadata + policy dry-run endpoint
- Org-level sharing + role-based assignments
- Tier-based quota enforcement + usage tracking
- Prometheus /metrics + real-time SIEM push (signed webhook)
- Compliance export API: SOC 2, EU AI Act, NIST AI RMF (stubs)
- GKE hardening: Binary Authorization ENFORCE, Cloud Armor WAF, Master Authorized Networks, NetworkPolicies, Secret Manager + CSI
- Public interactive demo + 4-tier pricing page
Next — design-partner track
- Terraform provider
- Native SDK adapters for LangChain, CrewAI, AutoGen, OpenAI Agents SDK
- Agent-to-agent auth (mTLS / token exchange)
- Advanced anomaly detection beyond shadow-agent heuristics
- Remaining 5th audit-logging phase
- Dedicated SIEM connectors (Splunk, Datadog) on top of webhook sink
- SOC 2 Type II external audit + ISO 27001 certification
- Self-hosted / private-cloud deployment profile
- SSO + SCIM provisioning
What We Believe
Three principles shape the product and the business.
Every AI agent should have a real, verifiable identity — not a shared key and a best guess.
Permissions should be explicit and least-privilege by default, with systems that fail closed when there's uncertainty.
Audit trails should be tamper-evident and useful for real investigations — not just another log table.
Let's talk
Whether you're exploring agent governance or ready to deploy, we'd love to hear from you. Design partner slots are open through Q2 2026.