Cyera's $1B Oasis Deal: NHI Is the Agent Era's Control Plane

By Deepak Bagada, CEO at SaaSNext & Principal AI Architect.

On July 28, 2026, Cyera agreed to acquire Oasis Security for roughly $1 billion — and with it, the entire identity industry acknowledged a simple fact: the explosion of AI agents has turned non-human identity (NHI) into the new control plane of enterprise security. Oasis Security is a specialist in NHI and Agentic Access Management (AAM): visibility into every service account, API key, token, and AI agent identity in your estate, plus the policy layer to control what those identities can do. It is the difference between "we know our humans" and "we know every machine identity that can reach production."

The deal is the biggest of the 2026 identity-consolidation wave, and it is surrounded by peers. CrowdStrike bought SGNL for $740M, Palo Alto Networks is paying $25B for CyberArk, and Cisco picked up Astrix for $400M. Cyera itself raised $600M at a $12B valuation to fund the Oasis deal, while Oasis had closed a $120M Series B in March 2026. The underlying statistic that explains all of it: non-human identities in the Fortune 500 grew roughly 500% in six months. Humans did not multiply fivefold; agents did. The latest AI news hub has been tracking this wave — here is the deep read on why machine identity is the agent era's control plane.

What non-human identity actually is

NHI covers every credential that is not a human password: service accounts, OAuth clients, API keys, workload identities, CI/CD tokens, and — new this year — agent identities. An AI agent that signs into your CRM, your GitHub org, and your payment processor is not "a user with a password." It is a machine identity with tokens, scopes, and a blast radius. In a typical enterprise, NHIs now outnumber human identities by 10x to 50x, and they are the identities attackers actually use: they never sleep, they rarely rotate, and they are chronically over-privileged.

Agentic Access Management: the Oasis thesis

Oasis's AAM product is built around three capabilities that make it the "control plane" of the agent era:

1. Visibility. Enumerate every non-human identity and the permissions it actually holds, including short-lived agent sessions, cloud workload identities, and SaaS API keys.
2. Control. Enforce least-privilege policies — auto-revoke unused keys, rotate long-lived credentials, and scope service accounts to the minimum surface.
3. Agent-to-agent auth. When one agent calls another (over A2A, MCP, or a vendor protocol), the identity doing the calling is an agent — AAM is what makes that handoff auditable and revocable.

Combined with Cyera's data-security platform (DSPM — discovering and classifying data everywhere), the acquisition narrative is clean: Cyera already knows where the data is; Oasis adds which machine can touch it.

The 2026 NHI acquisition wave

The pattern is unmistakable: the four biggest security vendors in the world each decided, within a single quarter, that machine identity is the next privilege-access market. Why now? Because agents compound the problem. One production agent orchestrating ten tools creates ten service accounts, ten API keys, and ten attack surfaces — and enterprises were watching NHI counts grow 500% in six months with zero governance tooling to show for it.

The identity-attack vectors that matter

Here is the dangerous compounding: a single over-privileged service account was a serious finding in 2024. Today, that same account is also a credential an AI agent holds, a credential that can be called on by a second agent, and a credential whose usage telemetry is invisible to the human security team. Each layer multiplies the blast radius instead of adding to it.

Practical code: find the over-privileged identities

The first step in NHI hygiene is enumeration. A minimal, provider-agnostic pattern — here shown for a cloud IAM audit — is to list every principal, expand each into the effective policies, and flag anything broader than the role's stated purpose:

import json

def audit_principals(principals, policies):
    findings = []
    for p in principals:
        effective = set()
        for role in p["roles"]:
            effective |= set(policies[role]["actions"])
        scoped = set(p["required_actions"])
        over = effective - scoped
        if over:
            findings.append({
                "principal": p["id"],
                "type": p["type"],  # service_account | agent | api_key
                "over_privileged_actions": sorted(over)[:10],
            })
    return findings

The result is the same shape a policy team needs regardless of cloud vendor: a list of machine identities whose effective permission set exceeds their job description. Run this on every service account before any agent is wired to it, and on every agent identity after an AAM rollout. The same discipline applies to the MCP servers your agents call — each tool connection is an identity boundary.

A least-privilege policy for agent identities

Once visibility exists, control is expressed as policy. A minimal JSON policy for an invoice-matching agent shows the shape:

{
  "identity": "agent:invoice-matcher",
  "policy_version": "1",
  "allowed_actions": [
    "billing:read", "payments:match", "reports:create"
  ],
  "denied_actions": [
    "payments:refund", "billing:delete", "iam:grant"
  ],
  "credential": {
    "type": "short_lived",
    "max_ttl_minutes": 60,
    "rotation": "automatic"
  },
  "agent_peers": ["agent:reconciliation"] 
}

Three properties in that file carry the whole security story: denied_actions (explicitly forbid the destructive ones), short_lived credentials (the token expires in 60 minutes regardless of whether the task finished), and agent_peers (only the reconciliation agent may call this one — the AAM equivalent of a firewall rule between agents).

What this means for security teams

• Machine identities are the new user accounts. If your security program still inventories humans and ignores service accounts, you are missing the identities that actually reach production.
• Agents need their own identity lifecycle. Provision, scope, rotate, revoke — the same lifecycle you apply to employees, applied to every agent that touches data. An agent without an identity lifecycle is an unmanaged insider.
• NHI governance is a prerequisite for safe AI workflows. Before you let an agent run a workflow across billing and CRM, you need to know which machine identities that workflow creates, what they can do, and who can revoke them.
• Expect consolidation to accelerate. With Cyera–Oasis, CrowdStrike–SGNL, Palo Alto–CyberArk, and Cisco–Astrix, the identity market is consolidating around four stacks. Procurement decisions made this year will lock in the vendor for the machine-identity era.

The control plane framing

Cyera's ~$1B Oasis deal is best read as a bet that identity is the control plane of the agent era. Agents act; someone must say what they are allowed to touch; that someone is no longer a human directory but a machine-identity plane with visibility, policy, and revocation. The 500% NHI growth rate guarantees the demand. The wave of billion-dollar acquisitions guarantees the supply. Non-human identity has gone from a niche audit checkbox to the fastest-moving category in enterprise security — and the agent era is only getting started.