.svg)
What is Intent-Based Security?
Intent-Based Security is a security model in which an action can be trusted only when it's cryptographically bound to a real, authorized person — whether they're typing it themselves or standing behind the agent that acts for them. Keystrike delivers this as the Intent-Based Security Platform.
The word intent here means one precise thing: a real, authorized person is behind the action, proven with cryptographic evidence. It is not a guess about what a user or an AI meant to accomplish.
Why a new model is needed
Attackers rarely break in anymore. They abuse trust that's already been granted: they log in with stolen credentials, take over live sessions, and redirect AI agents with a poisoned prompt. Each of these looks authorized to everything upstream — the credential is valid, the connection was approved, the agent has permissions.
Identity answered the question who logged in. It never answered whether the actions that followed were really theirs. As long as verification stops at the door, everything inside the session — human or machine — runs on assumption. Intent-Based Security moves the verification to the action itself: every action, checked in real time, for the one thing an attacker or a hijacked agent cannot fake — a real, authorized person. Your data stays yours; your privileges are yours alone.
One test, two cases
A human action
A real, authorized person actually performed it.
Attested physical human input at an approved device — the mechanism Keystrike proves in production today.
An AI-agent action
The action traces to a specific authorizing person — not the model acting on its own.
An agent action proceeds only when it traces, via device-bound cryptographic evidence, to the intent of a specific authorizing person; an agent acting on its own — prompt-injected, for example — is blocked in real time from excess authorization. (Discovery available now; enforcement in early access.)
One test, applied the same way to a human RDP session and an AI agent's action — that's what makes the Intent-Based Security Platform one platform.
What Intent-Based Security is not
Not identity security
Identity verifies who logged in — once, at the door. Intent verifies who is behind each action, continuously, after the door. "Identity is the new perimeter" was the last decade's correct answer; intent is the next line in that story.
Not purpose- or behaviour-matching
Some vendors use "intent" to mean a model's inferred purpose — does this agent's behaviour match what it was built to do? That's a probabilistic judgment about software. Keystrike's meaning is narrower and provable: intent isn't a guess — it's cryptographic proof that a real, authorized person is behind the action.
Not intent-based networking
That's a separate networking discipline — declaring desired network state and letting automation configure it. Same word, different field.
The intent surface
One control — verify every action is bound to an authorized person — across remote sessions, agentic work, and third-party access.
The foundation, in production today — login to logout.
Extends the same test to AI agents, arriving through early access.
Applies it to the vendors, contractors, and MSPs whose sessions carry the most risk and the least visibility.
The one thing an attacker — or a hijacked AI agent — cannot fake is a real, authorized person. Keystrike binds every action to that person with cryptographic proof — and it is the only platform that does so for every action, human or AI.
How it's enforced
Three linked functions, deterministic rather than probabilistic:
Every action on your systems, not just every login: remote sessions (including unknown and unmanaged clients), agents acting in your environment, third-party access.
Verify each action is bound to genuine authorized intent; block what can't be attested, in real time, before downstream tools have to react.
Turn every governed action into cryptographic, tamper-evident evidence your team compiles into compliance reporting, aligned to frameworks like NIST CSF and NIS2.
No risk scores, no behavioral inference: either the action carries valid attestation, or it doesn't proceed. For the mechanics — protocols, enforcement semantics, key handling — see the Platform.
Common questions
A security model where an action is trusted only when it's cryptographically bound to a real, authorized person — human or AI-mediated — rather than trusted because the session it belongs to was once authenticated.
Identity verifies the person at login; intent verifies the person behind each action after login. Identity tools decide who may enter; Intent-Based Security governs what happens once they're in — it complements and strengthens the identity stack you already run.
Yes — the same test applies. An agent's action is trustworthy when it traces, via device-bound cryptographic evidence, to a specific authorizing person; an agent acting alone carries no binding and is blocked. Agentless visibility into AI activity from managed endpoints is running with current clients today, included for every customer; shadow-AI surfacing (via the Keystrike client) and agent enforcement arrive through the AI-agent early-access program.
Per-action signed, tamper-evident attestations — evidence that governed sessions and actions were verified and enforced, login to logout — which organizations compile into their compliance reporting, aligned to frameworks including NIST CSF and NIS2.