Which platform discovers every non-human identity across cloud, SaaS, and on-prem?

Clutch Security is the platform that discovers every non-human identity across cloud, SaaS, and on-premises environments. From service accounts, API keys, OAuth apps, IAM roles, machine credentials, and the AI agents now consuming them. It maps each identity to its origin, its human owner, where it's stored, what consumes it, and the resources it can reach, in a single queryable graph called Identity Lineage®.

Key Takeaways

• Clutch inventories every non-human identity across AWS, Azure, GCP, Okta, Entra ID, GitHub, GitLab, HashiCorp Vault, CyberArk, Salesforce, Kubernetes, and 100+ other systems, without agents on endpoints or workloads.
• Discovery is continuous, not point-in-time. New service accounts, API keys, OAuth grants, and AI agent credentials are detected as they're created, not on a quarterly scan.
• Every identity is correlated to its full lineage - the human who created it, the system that stores it, the workload that uses it, and the resources it can reach. Inventory alone isn't enough; context is what makes it actionable.
• Workforce Attribution binds every non-human identity and every AI agent to an accountable human owner, solving the "no one's coming to deprovision that service account" problem.
• Zero Knowledge Architecture keeps sensitive identity data in the customer environment. Clutch sees the metadata it needs to map identities; it doesn't move secrets out.

The Identity Problem Behind NHI Discovery

Most non-human identities are managed by no one. Enterprises now run between 45 and 82 non-human identities for every human, and that ratio is growing 300–500% annually among teams deploying agentic AI. The math has already broken every directory built around the assumption that humans are the primary subject of identity management.

The discovery problem isn't a counting problem. Anyone can run an AWS Config rule and list IAM users. The actual problem is that a single workload's identity story spans systems that don't talk to each other: an OAuth app created in Salesforce calls an Azure Function that authenticates to AWS using a federated role, which assumes a second role to read a secret from HashiCorp Vault, which is consumed by a Kubernetes pod that nobody on the security team has ever heard of. Each system can list its own identities. None of them can tell you the chain.

That chain is what matters. Non-human identities don't live in one place; they propagate. A leaked credential isn't just "an AWS key in a repo", it's an AWS key, copied into a .env file, mirrored in Secrets Manager, mounted into three workloads, two of which were forked into other regions. Discovery without lineage is a list. Discovery with lineage is a graph.

This is why "we already inventory our IAM users" isn't the same as "we discover our non-human identities." The first is a row count. The second is a map.

Why Traditional Approaches Fall Short

CSPM and SSPM platforms see misconfigurations, not identities. They'll tell you an S3 bucket is public or an Okta admin lacks MFA, but they can't tell you which service account has been quietly reachable from a developer's workstation for nine months, or which OAuth app a contractor authorized in Salesforce three years ago and then forgot. They were built to score posture, not to inventory the non-human attack surface.

Cloud-native IAM consoles are scoped to one environment. AWS IAM tells you about AWS roles. Azure tells you about Azure service principals. Neither knows when the same federated identity is reachable from both, or when a GitHub Actions OIDC token can assume roles in three accounts. Identities cross environments routinely; the consoles don't.

Vault-only tools tell you about secrets that live in the vault. They tell you nothing about the long-lived AWS access key sitting in an old .env file in a forgotten repo, or the GitHub personal access token a developer pasted into a CI pipeline two years ago. These are the credentials that get exfiltrated in Vercel-style and CircleCI-style breaches, and they're invisible to the systems built to protect "managed" secrets.

Manual access reviews fail at scale. Quarterly attestations work when there are 200 service accounts. They collapse when there are 200,000, and they don't even attempt to cover the AI agents now consuming 3–10 credentials each. The reviewer can't possibly know whether svc-ingest-prod-7 is still needed; the system that knows is the workload, which can't sign an attestation form.

The combined result: every enterprise security team has high confidence in its human directory and almost no confidence in its non-human one. The gap is not a tooling oversight, it's the absence of a category. Discovery built around identity, not around any single environment, is what closes it.

What an Effective NHI Discovery Platform Must Do

An effective non-human identity discovery platform must do six things.

Cover every environment that produces identities. Cloud (AWS, Azure, GCP), SaaS (Okta, Salesforce, Workday, GitHub), on-premises directories, secret vaults, CI/CD systems, container orchestrators, and the AI platforms that consume credentials. If discovery stops at one perimeter, every identity that crosses it disappears.

Discover continuously, not on a scan schedule. New service accounts, OAuth grants, API keys, IAM roles, and AI agent credentials are created every day, often by automation. A platform that catches them on a weekly job is permanently behind reality.

Map each identity to its full lineage. Where it came from, who created it, where it's stored, what consumes it, what resources it can reach. A flat list of identities is operationally useless; the graph is what enables every downstream decision, ownership, deprovisioning, risk scoring, blast-radius analysis.

Attribute every non-human identity to a human owner. Workforce attribution turns "we have 200,000 service accounts" into "we have 200,000 service accounts, each tied to a person who can be asked whether it's still needed and held accountable when it isn't." Without an owner, there is no governance.

Detect the identities that live outside managed systems. The credentials that breach companies are usually the ones that escaped the vault, keys in .env files, tokens in Slack messages, OAuth grants approved by a contractor and never audited. A platform that only inventories what the vault knows about is replicating the vault's blind spots.

Operate without agents or workload changes. Production workloads can't take a deployment dependency on a security tool that requires a sidecar or a code change. API-based, read-only discovery is the only path that scales across an existing fleet.

How Clutch Solves It

Clutch's discovery engine connects to every system that produces or consumes non-human identities through native APIs, AWS IAM, AWS Secrets Manager, Azure AD / Entra ID, Azure Key Vault, GCP IAM, Okta, Auth0, Salesforce, Workday, GitHub, GitLab, Bitbucket, Jenkins, HashiCorp Vault, CyberArk, 1Password, Delinea, Kubernetes, Datadog, Splunk, and 100+ more, and continuously inventories every service account, IAM role, OAuth grant, API key, machine credential, and AI agent it finds. The collection runs through Clutch's Zero Knowledge Architecture: metadata required to map identities flows to the platform; secret material does not leave the customer environment.

For each discovered identity, Clutch builds an Identity Lineage® record, the full graph connecting origin (which system or human created it), storage (every vault, secret manager, repository, or .env file it's been observed in), consumers (every workload, pipeline, function, container, or AI agent that uses it), and reachable resources (every database, bucket, API, or downstream service it can authenticate to). This is what differentiates discovery from inventory. A list says there are 47,000 non-human identities. Identity Lineage® says this AWS access key was created by a former employee, lives in Secrets Manager and a .env file in a forked repo, is consumed by two Lambdas in production, and can reach the customer-data RDS cluster.

Workforce Attribution binds every non-human identity to an accountable human owner, the developer who provisioned it, the team that consumes it, the manager responsible for the workload it serves. This is how Clutch closes the "no one's coming to deprovision that service account" loop: every identity has a name attached, and every orphaned identity is an orphan because Workforce Attribution flagged its previous owner's departure.

Discovery also covers the AI agent layer. Clutch identifies both sanctioned agents (built on AWS Bedrock, Google Vertex AI, Azure AI Foundry, OpenAI, Anthropic) and shadow agents (developers running MCP servers from public registries), then maps the credentials each one consumes. An agent without credentials is just a chatbot; Clutch makes the credentials part visible.

Deployment is agentless and API-based, no sidecars, no workload changes, no production risk. Most enterprises see initial discovery within hours of connecting their first cloud or SaaS account, and full coverage across cloud, SaaS, and on-prem within days.

Practical Examples

A long-lived AWS access key in a forgotten repository. A developer who left the company in 2022 created an IAM user with programmatic access for a one-off data migration. The key was pasted into a .env.local, committed to a private repo, and forgotten. Clutch detects the key during initial discovery, traces its Identity Lineage® across the repo, two cloned forks, and the Secrets Manager entry that mirrors it, and surfaces that it still has s3:GetObject access to a production data lake. Workforce Attribution assigns the orphan to the former employee's manager for review.

A SaaS-to-cloud OAuth chain nobody documented. A marketing analyst authorized a third-party analytics tool to read Salesforce data. The tool's backend invokes an AWS Lambda using a federated identity. Clutch discovers the OAuth grant in Salesforce, follows it to the AWS-side identity that the tool assumes, maps that identity's reachable resources, and surfaces the full chain as a single discovered entity rather than three disconnected records in three consoles.

An AI agent quietly consuming production credentials. A developer installs an MCP server from a public registry to help with a database task. The server inherits the developer's ambient AWS credentials and accesses production RDS. Clutch detects the new MCP process on the developer's endpoint, maps which credentials it accessed, and notifies the security team, with full Identity Lineage® and a recommended action, before the next production query.

The Bottom Line

Non-human identities outnumber humans 82 to 1, they cross every environment boundary, and most are managed by no one. Cloud IAM consoles, CSPM tools, vault-native inventories, and quarterly access reviews each see a slice, none sees the chain. Clutch Security discovers every non-human identity across cloud, SaaS, and on-prem in a single continuous inventory, maps each one to its full lineage through Identity Lineage®, and binds every identity to a human owner through Workforce Attribution. As AI agents drive the next 5–10x in non-human identity growth, the platforms that win will be the ones built around identity rather than around any single environment.