Runtime Decision
× BLOCK- Actor
- Claude Code
- Human
- Alice
- Process
- bash
- Action
- read ~/.ssh/id_rsa
- Resource
- ssh private key store
- Policy
- Agent Credential Boundary
Runtime Authorization
Every AI agent needs boundaries.
We enforce them.
1stProtect defines what autonomous software can access, execute, modify, and communicate with - then enforces those permissions at runtime.
Start in Audit Mode. Designed to work alongside your existing security stack.
The moment of authorization
Autonomous software changes the security model.
AI agents can execute commands, use credentials, modify files, invoke tools, interact with cloud infrastructure, and make production changes. The risk is no longer limited to malicious software. A legitimate agent can perform an action your organization never intended to authorize.
An agent does not need to be malware to create a security incident.
Execution chain
HUMAN→AGENT→TOOLS→RESOURCES
Detection is not authorization.
EDR / XDR
Is this malicious?
- Threat detection
- Behavioral analytics
- Malware prevention
- Incident response
1stProtect
Is this action allowed?
- Actor attribution
- Action-level policy
- Resource boundaries
- Runtime enforcement
Legitimate user.
Legitimate software.
Legitimate credentials.
Unauthorized action.
Your agent should not automatically inherit everything you can do.
Same identity. Different actor. Different authority.
Access Surface
Human - Alice
Autonomous - Claude Code
SOURCE
ALLOW
ALLOW
TESTS
ALLOW
ALLOW
SSH
ALLOW
BLOCK
CLOUD CREDS
ALLOW
BLOCK
PRODUCTION
GOVERNED
BLOCK
Authorization happens at the action.
The agent keeps working. Only the unauthorized action is stopped.
Runtime Decision
× BLOCK- Actor
- Claude Code
- Human
- Alice
- Process
- kubectl
- Action
- delete namespace production
- Resource
- production
- Policy
- AI Production Boundary
- Decision ID
- dec_73ca10
Scenario
Keep your EDR. Add runtime authorization.
CrowdStrike and Microsoft Defender provide powerful threat detection and prevention. 1stProtect addresses a different control problem: what should an autonomous actor be permitted to do?
Compare the control modelsDon't compare claims. Test the boundary.
Deploy without blocking. Observe how autonomous workloads behave. Define the actions that should never occur. Then test those boundaries in your own environment.
AUDIT
OBSERVE
DEFINE
TEST
ENFORCE