AI Governance & Security Vendor Wiki

7 min read

Aembit

“Workload IAM” — secretless, policy-based, just-in-time access for workloads (and now AI agents) to the APIs, SaaS, and data they call, so credentials are never hardcoded or stored.

One-liner — A non-human identity platform that issues short-lived, policy-scoped credentials to workloads and AI agents at request time, so apps stop hardcoding secrets to reach the services they depend on.

Categories — non-human-identity

What it does

Aembit sells an access-management approach to non-human identity (NHI). Where most NHI vendors start with discovery and posture — find every service account, token, and API key and tell you which are risky — Aembit starts at the access decision itself. Its policy engine sits between a workload (a service, a container, a CI job, an AI agent) and the resource it wants to reach (a database, a SaaS API, another internal service), and it brokers that access: it cryptographically attests the calling workload’s identity, checks policy and posture, and then delivers a short-lived credential just-in-time, per task. The workload never stores or shares a long-lived secret.

The pitch is that this removes the root cause of most NHI incidents — hardcoded keys, sprawling service-account secrets, over-permissioned tokens that never rotate — rather than just inventorying them after the fact. Capabilities include workload attestation, ephemeral/secretless credential issuance, least-privilege policy, and centralized auditing of every machine-to-machine access.

Where it sits in the stack

Primary category: non-human-identity (Foundation layer). Aembit is the “access management” pole of NHI, as distinct from the “posture/discovery/governance” pole occupied by oasis-security, token-security, astrix-security, and clutch-security. By gating which workload/agent can reach which resource, with short-lived scoped credentials, it constrains an agent’s ability to reach sensitive data and to exfiltrate via over-broad service-to-service access. It does not inspect prompts or untrusted input. It lives at the boundary between trust zones — it is the broker that decides whether a workload in one zone may call a resource in another.

Deployment & architecture

  • SaaS control plane — Aembit’s policy/credential service is delivered as a managed cloud service; for agentic AI this includes an MCP Authorization Server implementing the OAuth 2.1 authorization-code flow and integrating with existing IdPs (Okta, Microsoft Entra/Azure AD, Google).
  • Edge enforcement component — a customer-deployed gateway/agent (for the agentic product, an MCP Identity Gateway that runs as a Linux VM) validates tokens and enforces policy on every request, performing real-time credential exchange so the workload/agent “never holds direct credentials.” For classic workloads, Aembit deploys an edge agent (e.g., sidecar/host) that intercepts the workload’s outbound calls and injects the just-in-time credential.
  • Integrations — IdPs for the human side of identity, secrets managers, and the target SaaS/API surfaces; for agents, it secures Claude and Microsoft Copilot Studio agents over MCP.
  • Blended Identity — its access model evaluates the AI agent’s identity and the human operating it together in a single policy decision at request time (“is this user, via this agent, allowed to reach this resource right now?”), with per-user credential isolation per session.

Positioning & differentiators

Aembit’s differentiator is that it is enforcement-first, not inventory-first. oasis-security, token-security, astrix-security, and clutch-security largely lead with discovering NHIs, mapping their entitlements and ownership, scoring risk, and helping rotate/clean up — a posture and governance motion that overlays existing identity providers and secrets stores. Aembit instead inserts itself in the access path and brokers the credential, aiming to make long-lived secrets unnecessary in the first place. That makes its closest conceptual neighbor the workload-identity / machine-auth lineage (SPIFFE/SPIRE, and the secretless side of cyberark’s machine-identity / Conjur portfolio) rather than the NHI posture startups. It also moved early and visibly on agent identity / MCP, shipping GA “IAM for Agentic AI” in April 2026 and MCP governance for Copilot Studio agents at Identiverse in June 2026.

Tradeoff: enforcement-first means deploying Aembit in the access path (edge gateway/agent) and modeling policies, which is more invasive than read-only posture tools — higher value if you adopt it, more integration work to stand up.

Ownership, funding & M&A

Independent and venture-backed. Founded 2021, HQ Silver Spring, Maryland. Co-founders David Goldschlag (CEO) and Kevin Sapp, who previously built New Edge Labs (acquired by Netskope), an early user-side zero-trust product; their backgrounds span the identity/zero-trust space (the brief’s specific Zscaler/Mandiant tenure was not corroborated in the primary sources reviewed — treat those specifics as unverified). Funding: $25M Series A on 2024-09-12 led by Acrew Capital, with Ballistic Ventures, Ten Eleven Ventures, Okta Ventures, and CrowdStrike Falcon Fund participating, bringing cumulative funding to ~$45M (vendor + SecurityWeek). The earlier seed (~$16.5M) rolls into that ~$45M cumulative figure. No acquisition found — the seed registry carried no M&A flag and research found none; ownership independent, confidence high (corroborated by independent press). Note the investor base (Okta Ventures, CrowdStrike Falcon Fund) ties it to two identity/security incumbents — worth watching as NHI consolidates.

CTO / hedge-fund lens

This is Day-2 for most shops and, for a typical hedge fund, low fit today. Aembit solves a problem that bites cloud-native engineering organizations with lots of workload-to-workload and workload-to-SaaS traffic and a sprawl of service-account secrets — i.e., teams that build and run substantial software. A 50-person fund running mostly SaaS and a couple of internal apps gets more NHI risk reduction sooner from secrets hygiene in its existing vault and IdP than from standing up a workload-IAM broker. It has no direct SR 11-7 / model-risk role; it is an access-control control, not a governance or model-risk tool. Where it becomes interesting for a fund is the agentic-AI angle: if you start running AI agents that autonomously call internal APIs and SaaS (via MCP), Aembit’s Blended Identity and per-session short-lived credentials are a credible way to keep agents from holding standing secrets or acting beyond the operating human’s entitlements — a control you will want before agents touch anything sensitive. Until you are at that stage, this is a watch-list item, not a Day-1 buy.

Competitors / alternatives

oasis-security, token-security, astrix-security, clutch-security, cyberark (machine identity / Conjur), entro-security, natoma; conceptually also SPIFFE/SPIRE for open-source workload identity.

Open questions / to verify

  • Founders’ exact prior employers (brief cites ex-Zscaler/Mandiant; only New Edge Labs→Netskope was corroborated here).
  • Seed round size/date and lead (brief notes ~$16.5M; not separately confirmed in cached sources).
  • Any funding round after the 2024-09 Series A (none found as of 2026-06).
  • Production traction/customer count beyond named references (Snowflake, Fortune 250 retailer).
  • How heavy the edge-agent deployment is in practice for classic (non-agent) workloads.

Sources

History

  • [2026-06-28] Stub created from seed registry.
  • [2026-06-28] Researched; established founded 2021, HQ Silver Spring MD, founders David Goldschlag (CEO) & Kevin Sapp (ex-New Edge Labs/Netskope), $25M Series A 2024-09 led by Acrew (~$45M cumulative), independent (no M&A). Documented secretless/JIT workload-IAM access-management model (vs posture/discovery NHI peers), SaaS control plane + edge gateway/agent architecture, and the agentic-AI/MCP push (Blended Identity, MCP Authorization Server + Identity Gateway, Claude/Copilot Studio, April-2026 GA). Set ownership_confidence high, hedge_fund_fit low, status researched, confidence medium. 4 sources cached.

Graph View

Backlinks