Your AI agents act first. You find out later. Oathera checks every AI agent before it acts, with machine-bound non-human identity (NHI) control and governance for autonomous AI agents.
Borrowed key
Nothing checks your AI agents before they act. A personal agent carries everything its user can do, and a stolen credential works for whoever holds it. Your identity provider decided who could sign in; nothing checks each action before it runs.
Today's agent
- ID
- agt-4f81c2
- Name
- demo-agent
- Secret
- ••••••••••••
Nothing more. Whoever holds the secret is the agent.
Oathera
Before a call reaches your tool, Oathera knows which agent is making it, for whom, and whether this action is one it may take: allowed to read, not to write; no agent acts as admin.
Oathera gives every AI agent an identity bound to the machine it runs on, with a named person who answers for it, and checked every time it acts.
Your agent with Oathera oathera:demo-agent.23aba0ed8d80@ais.oathera.org
What it claimedAn autonomous agent for reports
Enrollment attribute block: agent type autonomous, capabilities report-generation and data-analysis, OS Linux, version 1.0.0.
What Oathera found on its hostHost digest MYQAQXgb9kC3hqXx
Host confidence level 3, where 1 is a verified host. There is no platform attestation, so there is one finding: the host details come from the agent’s own process.
The person who approved itJohn Scully
Principal: john@northwind.example, email verified, signed in at the organization’s identity provider.
The organizationGroups engineering and platform
From the organization’s identity provider at sign-in: groups engineering and platform.
And on every call, your agent's fingerprint: a single-use proof signed by a key held only in its memory.
From an enrollment on Oathera's demo, 16 Oct 2026. John Scully is a demo user.
Borrowed key
One key, copied into every agent.
You can't tell which agent did what, and you can't stop one without breaking a person's account.
Oathera
Enroll. Check every time. Prove every call.
- Enroll Oathera Agent Enrollment, on your MCP server, enrolls the agent: its own key, created on its machine, and the host. A person approves it before it acts, through your identity provider or the Oathera console.
- Check every time Each time the agent asks for a token, the Oathera Identity Service compares the host with the enrollment record and refuses a mismatch.
- Prove every call Your agent's fingerprint is on everything it does. Oathera Agent Enforcement in front of your tools verifies each signed operation and asks the Oathera Identity Service for a decision before the call goes through.
Borrowed key · An agent with no identity
An agent calls your tools with a borrowed key.
The call goes through. Nobody can say which agent made it.
Accepted
Oathera · An agent with no identity
No identity, no call.
Oathera Agent Enforcement refuses the call before it reaches the tool, and the console lists the agent so you can enroll it or reject it.
Refused · invalid_token
Borrowed key · An identity of its own
A shared key can't tell agents apart.
It works for anyone who holds it, from any machine, until someone rotates it.
Works for anyone
Oathera · An identity of its own
Its own key, approved by a named person.
Its identity token lasts five minutes at most, every call carries a single-use proof, and your service gets a gateway assertion in place of the token.
Checked · proof verified
Borrowed key · Impersonation
Another agent presents the same key.
The service can't tell them apart, so it answers both.
Accepted
Oathera · Impersonation
A token on its own is worth nothing.
Another agent presents this agent's token. It can't prove it holds the session key, so the call is refused.
Refused · the proof fails
Borrowed key · Cloning
Copy the agent to another machine.
The copy works just like the original.
Accepted
Oathera · Cloning
Same code. Different machine. Different fingerprint.
The copy is caught at its next token: its host is compared with the enrollment record. Under enforce it is refused; under warn it is flagged with a high-severity alert.
403 host_changed
Borrowed key · Outside its boundary
The agent reaches for a tool it wasn't set up for.
The key opens whatever it was scoped to.
Accepted
Oathera · Outside its boundary
Outside its boundary, refused.
An unknown operation, a tool the server doesn't list, arguments outside the allow-list, an assertion made for another service: each one is refused.
Refused · unknown_tool
Borrowed key
Every time, the key said yes.
Oathera
Every time, the agent had to prove it.
Every AI agent, sworn in.
An identity bound to the machine it runs on, with a named person who answers for it, and checked every time it acts.
We start with a small group of design partners in early November 2026: a dedicated tenant, direct access to the founders, and a say in the roadmap. Pricing confirmed at launch.
- A copied agent is refused. The agent's key is created on its own host. The host is recorded at enrollment and compared again at every token.
- A stolen token authorizes nothing. Tokens last five minutes at most, and every operation carries a proof signed with a key that never leaves the agent's memory.
- A person answers for every agent. A person approves each agent before it acts, through your identity provider or the Oathera console. Deprovision a person and all of their agents stop, without touching their own sign-in.
Built for platform and AI teams that run internal MCP servers, and the security leaders who answer for them. Oathera fits beside your identity provider, your egress controls and your AI governance. It doesn't judge model output, inspect or rewrite prompts, replace your identity provider, or verify people itself.
The five use casesOne agent, start to finishThe data sheetsWatch the attacks fail in the simulator
Two questions about every agent.
What did my agent do? Does my agent have its DNA? Oathera answers both, for every AI agent you run.
What did my agent do?
Every call your agent makes carries its fingerprint: a proof signed for that one operation, with a key held only in the agent's memory. Oathera Agent Enforcement checks it before the call goes through, and every decision is recorded, so each action traces to one agent and the person who answers for it.
Does my agent have its DNA?
Oathera records your agent's DNA once, when it enrolls: its own key, created on its own machine, the machine it runs on, and the person who answers for it. A copy on another machine does not match it, and gets no token.
An identity of its own, for every agent.
Oathera gives every AI agent an identity bound to the machine it runs on, with a named person who answers for it, and checked every time it acts.
Bound to its machine.
Copy an agent and you get the same code with a different fingerprint. The agent's key is created on its own host, the host is recorded when the agent enrolls, and it is compared again every time the agent asks for a token.
Accountable.
Every action traces to an agent and the person who answers for it. A person approves each agent before it acts, through your identity provider or the Oathera console, and the approval is part of the agent's enrollment record.
Fits what you run.
Works with your identity provider. We don't replace your IAM or AI governance; we're the identity layer under them. People approve agents by signing in through your identity provider over OIDC.
Enroll. Check every time. Prove every call.
The Oathera MCP Plugin works in your environment, on your MCP server: Oathera Agent Enrollment and Oathera Agent Enforcement. The Oathera Witness runs on each machine where agents run. The Oathera Identity Service keeps the record.
1. Enroll
Oathera Agent Enrollment, on your MCP server, enrolls the agent: its own key, created on its machine, and the host. A person approves it before it acts, through your identity provider or the Oathera console.
2. Check every time
Each time the agent asks for a token, the Oathera Identity Service compares the host with the enrollment record and refuses a mismatch.
3. Prove every call
Oathera Agent Enforcement runs in your MCP server and verifies each signed operation. For an action that changes something, it asks the Oathera Identity Service for a decision before the call goes through.
Three Oathera components. Everything else stays yours.
Your identity provider, your agents and the machines they run on, your MCP servers, your policy engine and your services stay yours. Oathera works with them and replaces none of them.
Oathera Witness
On every machine where agents run, your servers and laptops. It watches the agents there and vouches for what it saw in a signed report.
Oathera MCP Plugin
With your MCP server. Oathera Agent Enrollment gives each agent its identity; Oathera Agent Enforcement checks every call.
Oathera Identity Service
Oathera's cloud. It keeps the record of every agent, issues its identity tokens and records every decision.
Built for the teams that put agents to work.
Platform and AI teams
The first users: a platform or AI-enablement team that runs internal MCP servers and wants each agent that calls them to prove who it is.
Security leaders
The buyer: the security leader who has to answer which agent did what, and stop one without breaking anything else. Developers adopt it; the security budget pays.
Beside what you run
Oathera fits beside your identity provider, your egress controls and your AI governance, not instead of them. It doesn't judge model output, rewrite prompts, replace your identity provider, or verify people itself.
Become a design partner.
We start with a small group of design partners in early November 2026: a dedicated tenant, direct access to the founders, and a say in the roadmap. Pricing confirmed at launch.