The most important user in an enterprise may no longer have a face, a password or the patience to wait for an approval prompt. It could be a payroll script querying a database, a GitHub job shipping code, or an AI agent opening a financial model through an MCP server. These machine actors work continuously. Many still get into sensitive systems with credentials copied into environment variables, configuration files and service accounts - keys that can linger long after the task that justified them.
Aembit is building identity and access management for that population. The Silver Spring, Maryland company places a control plane between software that wants access and the service it wants to reach. It verifies the workload, evaluates context and policy, and supplies a short-lived credential when the request passes. The client application need not store the credential or even know it. Security teams get a decision record; developers avoid writing bespoke authentication plumbing.
The key that became the problem
Founders David Goldschlag and Kevin Sapp did not arrive here by discovering that passwords are bad. They had already spent years on identity and zero trust, including founding New Edge Labs, an early zero-trust network access company acquired by Netskope in 2019. Goldschlag's résumé reaches further back: at the National Security Agency, he co-invented onion routing, the technical foundation of Tor. The two had worked together for nearly two decades when they began talking with previous customers about a less mature frontier.
What failed first was the handoff between applications. Conventional secrets vaults helped store credentials, but storage did not answer whether a particular workload, in its present condition, should receive access. Cloud IAM worked inside its own boundaries, then became awkward when an AWS job needed Azure data, an on-premises server needed SaaS, or a pipeline needed a third-party API. Aembit says those customer conversations changed the frame. The founders stopped treating machine access as a secret-distribution chore and began treating it like IAM.
“We see ourselves in our customers because we have been in their shoes.”David Goldschlag, co-founder and CEO
Aembit formally launched in March 2023 with $16.6 million in seed financing. Acrew Capital led a $25 million Series A announced in September 2024, joined by Ballistic Ventures, Ten Eleven Ventures, Okta Ventures and CrowdStrike Falcon Fund. The round brought disclosed capital to nearly $45 million. The money bought runway in a category with an ungainly name - non-human identity - but a clean economic premise: software now outnumbers the humans an identity team manages, and every automated connection can create standing privilege.
What happens in the milliseconds before yes
The product begins with a client workload - an application, script, CI/CD job or agent asking for something. A trust provider establishes an identity using evidence from systems such as Kubernetes, cloud IAM, OIDC, SPIFFE or a development platform. Aembit's policy engine then checks who the client is, what resource it wants and, where configured, conditions such as workload posture, location or time. Approval triggers a credential or token in real time. Aembit Edge can inject it into the network request, keeping it out of the application's memory and configuration.
and context
access policy
credential + log
This is the distinction between Aembit and a conventional secrets manager. A vault is useful when a company needs a controlled place to keep and rotate a password. Aembit wants the client to avoid possessing that durable password in the first place. It is also different from discovery-led non-human identity products such as Oasis, Token Security, Entro or Astrix, which start by inventorying service accounts, keys and OAuth grants. Aembit can discover relationships, but its sharper edge is enforcement in the request path.
The alternatives are not always competitors. A company may use a discovery product to learn what exists, a vault for legacy secrets, cloud-native federation where it fits, and Aembit for access that crosses environments. Pomerium and Britive overlap more directly on runtime control; CyberArk and Akeyless approach from privileged access and secrets; a sophisticated platform team can build with SPIFFE and OIDC. Aembit's commercial argument is that stitching those pieces together becomes its own product, with an audit and maintenance bill attached.
A customer result with actual arithmetic
Snowflake is Aembit's most legible public customer. Its security team described inconsistent access methods, persistent secret risk, manual rotation and weak dynamic control. It started by securing access from several workloads to an internal Snowflake instance. In Aembit's published case study, Snowflake reports that secretless, identity-based access cut credential issuance, rotation and audit follow-up by 85 percent. As deployment expands, the team estimates savings of five to ten hours a day across credential management, reviews, reporting and process work - roughly two full-time roles.
Other case studies show the wedge. A Fortune 250 retailer used Aembit for access to HashiCorp Vault after a homegrown identity system became difficult to scale. A global property-management firm removed secrets from a configuration system without first standing up a separate secrets manager. Red Cup IT uses the product around AI agents deployed into customer environments. An unnamed investment firm with about $300 billion under management put Claude agents behind policy, logging actions into CrowdStrike's SIEM. Aembit says that deployment took two weeks and less than six hours of the security team's time.
Then the machines started speaking English
AI agents sharpened the identity problem because one action can carry two identities. There is the agent calling a tool and the employee on whose behalf it acts. A shared service account flattens that context. Aembit's “blended identity” keeps the user and agent together, while its MCP Identity Gateway enforces policy, exchanges tokens and logs tool calls. The ambition is to answer four questions after an event: which agent acted, for which user, on what resource and under what restriction.
Aembit introduced the agentic product in October 2025 and made it generally available in April 2026 after customer alpha and beta work. Subsequent releases added Microsoft Copilot Studio, an Aembit MCP server for querying access activity, Oracle database support and secretless Azure Databricks pipelines. In July 2026, Snowflake and Aembit announced an integration aimed at governing third-party agents reaching Snowflake data while preserving user-and-agent attribution.
“Aembit can save us five to 10 hours a day.”Cameron Tekiyeh, Snowflake, on the expanding workload deployment
What it costs - and what you can copy
Aembit's published pricing is unusually plain for enterprise security. The free workload plan covers up to 10 workloads and 10 access policies; the agent plan covers three agents, one MCP Identity Gateway and five authorization policies. Team plans start at $20 per workload or agent each month, with ceilings of 50 workloads or 500 agents on the published page. Enterprise contracts add unlimited scale, conditional access, custom log retention and around-the-clock support. The real cost includes deployment, policy design and having another control plane in a sensitive path.
The pilot pattern worth stealing
- Choose one painful route - a CI job to cloud API, not the whole estate.
- Prove the client identity from a platform signal it already has.
- Replace one long-lived secret with a short-lived, narrowly scoped token.
- Put policy and an audit event around every request.
- Measure rotation work, failed access and developer changes before expanding.
That sequence is useful even without buying Aembit. The copyable insight is to start from a connection, not from an inventory spreadsheet. Pick a route whose credential is frequently rotated or broadly shared. Establish identity, narrow the permission and instrument the decision. A clean pilot should show whether the organization is removing work or merely moving it into a new console.
Where the badge does not fit
Secretless access is not a universal solvent. It works when the workload has reliable evidence of identity, the destination accepts a credential that can be issued or brokered, and the network path can tolerate an enforcement component. A legacy application with a fixed password field may still need a vault. A small system contained in one cloud may already get enough from native workload identity federation. A team that cannot name its machine identities may need discovery before enforcement.
There is also concentration risk. Putting policy in the request path makes that service operationally important, even when edge components are designed for resilience. Bad identity evidence produces confident bad decisions; bad policy can block production or grant too much. And per-workload pricing becomes material when ephemeral jobs multiply. Buyers should test failure behavior, latency, offline tolerance, credential compatibility and how cleanly they can exit.
Aembit nevertheless occupies a useful market seam. Human IAM proved that identity could be a control plane rather than a password list. The company is applying that lesson to the automated systems now carrying enterprise work - first services, then pipelines, now agents. Its best product demo is not a dashboard. It is the missing secret: the credential that appeared for one approved task, lived briefly, and left behind enough context to explain exactly why the machine got in.