BEFORE ACTION, AUTHORITY.

Agents need more than access.

Praeva is runtime authority infrastructure for AI agents and the independent Authority Control Point (ACP) between agentic intent and consequential execution. It verifies that an agent has legitimate, current, and properly delegated authority before the action executes.

Valid credentials are not proof of valid authority.
Independent by design Delegation lineage Authority receipts
LIVE AUTHORITY PATH EVALUATING
01 PRINCIPAL Human or system
02 AI AGENT Requests action
P
PRAEVA Independent Authority Control Point
03 TOOL / SYSTEM Execution boundary
DECISION
ALLOW
DENY
ESCALATE
Authority lineage verified · scope current · constraints satisfied
THREE DIFFERENT QUESTIONS

Identity and access get an agent to the door. Authority determines whether the action should cross it.

01
IDENTITY Who is acting?

Authentication establishes the actor.

Establish
02
ACCESS What can it reach?

Credentials unlock systems and tools.

Reach
Praeva complements identity and access controls. It evaluates the authority behind a consequential action before execution.
THE MISSING CONTROL LAYER

Authentication can remain valid after authority has failed.

AI agents can hold valid credentials, call approved tools, and still move beyond the purpose, scope, time, or delegation originally granted to them.

Praeva evaluates the authority behind the action—not merely whether access exists.

01

Identity

Establishes who or what the actor is.

“Is this really the agent?”
02

Access

Determines what a credential can technically invoke.

“Can this credential reach the system?”
WHY NOW

Agents are moving from generating output to taking action.

As AI agents gain access to enterprise tools, APIs, systems, and data, the security question changes. It is no longer enough to ask whether an agent can authenticate or whether a credential has permission.

Consequential execution requires a second question: does the agent still possess legitimate authority for this action?

01

More autonomy

Agents increasingly initiate workflows, call tools, and make changes without a human at every step.

02

More delegation

Authority passes from people and systems into agents, then across tools, services, and downstream agents.

03

More consequence

The same credentials that enable useful automation can also enable actions well beyond the original purpose.

WHERE PRAEVA FITS

Praeva is the independent Authority Control Point.

Praeva sits outside the acting agent’s trust boundary and evaluates delegated authority before consequential execution. Independent describes the control boundary, not who built the technology.

CONTROL PRIMARY QUESTION PRAEVA RELATIONSHIP
IAM / Authentication Who or what is this? Praeva consumes identity context; it does not replace it.
Authorization / Policy What can this credential or role invoke? Praeva evaluates whether the requested action remains within legitimate delegated authority.
EDR / Detection What suspicious behavior is occurring? Praeva acts before consequential execution rather than only detecting behavior afterward.
Praeva Does this agent have legitimate, current authority to do this now? Independent pre-execution authority verification + decision evidence.
INDEPENDENT BY DESIGN

The actor should not judge its own authority.

The system taking the action should not be the system that determines whether it still has authority to take it. Praeva provides an independent Authority Control Point before execution.

Independent describes the control boundary, not the vendor. An enterprise may run the ACP itself or consume it as infrastructure. What matters is that the acting agent cannot self-approve, bypass, or rewrite the authority decision.

01

Outside the actor’s trust boundary

Praeva evaluates authority independently of the agent, model, or orchestration stack requesting execution.

02

Built by anyone

Model, cloud, framework, and agent vendor do not determine authority. The same independent check applies to every agent.

03

Before execution

The ACP returns ALLOW, DENY, or ESCALATE before the target action proceeds, then preserves an Authority Receipt.

We do not care who built the agent. We care whether it has authority to act.
HOW PRAEVA WORKS

Authority becomes enforceable at runtime.

Praeva sits in the execution path as an independent Authority Control Point between an agent and a consequential downstream action. The ACP verifies authority before execution and records why each decision was made.

REFERENCE EXECUTION PATH PRE-EXECUTION CONTROL
01 Principal Originates authority
delegates
02 AI Agent Requests action
requests
03 Praeva ACP Independently verifies authority
decides
04 MCP / REST / Tool Executes action
ACCESS CREDENTIAL May still be valid
≠
RUNTIME AUTHORITY Must still be legitimate
DEFINE
01

Authority Registry

Principals, agents, objectives, actions, constraints, expiry, and revocation.

REPRESENT
02

Delegation Credential

Short-lived authority represented separately from the target system credential.

PROVE
04

Authority Receipt

Preserves decision evidence, lineage, requested action, resource, and rationale.

RUNTIME SCENARIO

The credential remained valid. The authority did not.

A legitimate workflow agent is compromised. Its credential still works. Praeva evaluates each attempted action against the authority actually delegated.

Authenticated ≠ Authorized by authority
AGENT EXECUTION TRACE TRACE 8F2A17
00:01
Inspect approved configuration Delegated purpose + scope verified
ALLOW
00:07
Retrieve unrelated secrets Outside delegated objective
DENY
00:09
Pivot into another system No valid authority chain for target
DENY
00:12
Create persistent administrator Privileged action requires approval
ESCALATE
00:15
Bulk encrypt or destroy data Consequential action exceeds authority
DENY

Prevention first. Evidence always.

A perfect audit trail cannot undo an unauthorized action.

Most systems can tell you what an AI agent did. Praeva is designed to determine whether an agent has legitimate authority to do it before the action reaches execution.

After the action

Recording what happened

Logs and decision records help reconstruct actions later. Evidence alone cannot prevent an unauthorized action.

Before the action

Controlling what happens

Praeva checks current delegated authority at the protected execution boundary and returns ALLOW, DENY, or ESCALATE before proceeding.

When required authority evidence is missing, invalid, expired, or revoked, the protected action does not proceed.

Each evaluation also produces an Authority Receipt: what was requested, what Praeva decided, why, and what enforcement or downstream execution evidence is actually available. Evidence not obtained is explicitly identified as unverified. Enforcement guarantees apply to execution paths integrated with the Praeva control point.
DECISION EVIDENCE

First enforce the boundary. Then prove what happened.

Every consequential decision can preserve the context needed to explain who delegated authority, what action was attempted, which constraints applied, and why Praeva allowed, denied, or escalated it.

  • Originating principal
  • Delegated purpose
  • Requested action + resource
  • Authority lineage + constraints
  • Decision + rationale
PRAEVA AUTHORITY RECEIPT PRV-8F2A17
ESCALATE
Principal
FINANCE_OPS
Agent
AP_AGENT_04
Action
CREATE_VENDOR_ADMIN
Purpose
Quarter-end reconciliation
Authority state
OUT_OF_SCOPE
Delegation
VERIFIED
AUTHORITY DOCTRINE

Authority should narrow as work moves outward.

Praeva is built around a simple idea: an AI model should not be able to invent, widen, or ratify its own authority.

01

Authority originates outside the model.

Legitimate authority comes from a human, organization, system, or other recognized source—not from model confidence.

02

Delegation should be explicit.

Purpose, scope, actions, resources, constraints, expiry, and lineage should be representable and inspectable.

03

Authority should be independently verifiable.

The system taking the action should not be the system that decides whether the action remains within valid delegated authority.

04

Authority must be revocable.

Delegation can expire, be withdrawn, or become invalid as context changes. Runtime decisions must reflect that change.

05

Decisions should leave evidence.

Organizations should be able to reconstruct who delegated authority, what was requested, and why the action was allowed, denied, or escalated.

PRAEVA INSIGHTS

When Agents Run Amok.

A continuing Praeva Systems series examining agent failures through one specific lens: where did legitimate authority end?

View all resources →
WHEN AGENTS RUN AMOK#010
NEW · November 12, 2026

United Said No to AI Agents. The Real Question Is Who Gets to Say Yes.

United's restrictions on unauthorized third-party AI agents raise a fundamental question: who can verify what an agent is legitimately authorized to do for a customer?

Read #010 →
PRAEVA / AUTHORITY BEFORE ACTION CUSTOMER DELEGATION What did the customer approve? PLATFORM PERMISSION Is this channel permitted? VERIFY AUTHORITY BEFORE EXECUTION
WHEN AGENTS RUN AMOK #009
October 31, 2026

We Built the Guardrails. Then We Tried to Break Them.

The founder tests stopped unauthorized transactions. But can an agent bypass the gateway, or can the evidence be altered? The next challenge is independent enforcement-path verification.

Read #009 →
WHEN AGENTS RUN AMOK #008
NEW · OCTOBER 16, 2026

We Turned the Orcas Loose

We built a working authority control point, challenged it with 20 real AI-model sessions, and tested what happened when five in-flight transactions lost their parent authority. The founder POC results, with the limits clearly stated.

Read #008 →
WHEN AGENTS RUN AMOK #007
OCTOBER 8, 2026

Kill Switch vs. Stop Sign

Kill switches matter. But sometimes the agent is fine and the action is not. #007 looks at the difference between shutting an agent down and stopping the wrong action before execution.

Read #007 →
When Agents Run Amok #007: Kill Switch vs. Stop Sign
WHEN AGENTS RUN AMOK #006
SEPTEMBER 23, 2026

The Work Was Authorized. Then It Wasn't.

The worker, credential, and queued operation can remain unchanged while the originating authority disappears. #006 asks why execution-time revalidation matters.

Read #006 →
When Agents Run Amok series graphic
WHEN AGENTS
RUN AMOK
WHEN AGENTS RUN AMOK #005
NEW · SEPTEMBER 16, 2026

An Agent Should Never Be Able to Approve Itself

An approval artifact is not proof of authority merely because it exists. #005 asks whether the actor requesting power can also be trusted to create the evidence that grants it.

Read #005 →
When Agents Run Amok series graphic
WHEN AGENTS
RUN AMOK
WHEN AGENTS RUN AMOK #004
NEW · SEPTEMBER 3, 2026

The Problem With Proving an Agent Was Wrong

Logs can reconstruct what an agent did after the fact. The harder question is whether the agent had legitimate, current, delegated authority before the action occurred.

Read #004 →
When Agents Run Amok series graphic
WHEN AGENTS
RUN AMOK
WHEN AGENTS RUN AMOK #002
NEW · AUGUST 26, 2026

Who actually authorized the agent?

An agent can be authenticated, hold valid credentials, and still take an action nobody actually authorized. The second installment looks at why identity and access stop short of answering the authority question.

Read #002 →
When Agents Run Amok series graphic
WHEN AGENTS
RUN AMOK
WHEN AGENTS RUN AMOK #001
INCIDENT ANALYSIS

The credential remained valid. The authority did not.

A compromised agent can remain authenticated while its actions move far beyond the purpose originally delegated. What should the execution boundary check?

Read #001 →
PRIMER

Identity, access, and authority are not the same thing.

A concise explanation of the control gap Praeva is designed to address.

Read primer →
DOCTRINE

Five principles for delegated agent authority.

The public principles that guide Praeva’s architecture and product decisions.

Read doctrine →
ABOUT PRAEVA

Serious infrastructure. Built by people who give a damn.

Praeva Systems is building runtime authority infrastructure for AI agents. At its center is an independent Authority Control Point (ACP) that verifies delegated authority before consequential execution. Our technology is designed around a simple principle: consequential power should come with clear authority, defined limits, and accountability.

We think companies should work the same way. Praeva is being built as a place where excellent work and decent treatment of people are not competing ideas. We expect high standards, real accountability, and respect for the people doing the work.

You should not have to become less yourself to become more successful here.

We do not change our values to suit the politics of a ZIP code.

Praeva will follow the law. We will also advocate for our people and refuse to voluntarily adopt exclusionary practices simply because doing so would be easier. Compliance is a floor. It is not our moral ceiling.

THE PRAEVA PEOPLE DOCTRINE

Real values should show up in how people are treated.

These are not slogans for a recruiting page. They are the standards we intend to use when we make decisions about people, benefits, leadership, opportunity, and culture.

01 · DIGNITY

Dignity is non-negotiable.

Race, color, national origin, sex, gender, gender identity or expression, sexual orientation, religion, disability, age, family structure, veteran status, pregnancy, neurodiversity, or background do not diminish anyone's place here.

02 · BELONGING

Belonging does not mean conformity.

Employees are not required to share the founder's politics, vote a particular way, or agree with every position Praeva takes. They are required to treat other people with respect.

03 · INCLUSION

Inclusion is something we do.

Our values should show up in who gets hired, who gets heard, who gets promoted, how people are paid, how benefits are designed, and whether people can fully participate in the company.

04 · FAMILY

Family means family.

Spouses, partners, children, adoptive and blended families, caregivers, and the many ways people take care of one another deserve policies that reflect real life.

05 · OPPORTUNITY

Talent is widely distributed. Opportunity is not.

We will work to build recruiting, compensation, promotion, mentorship, and development practices that give talented people a genuine opportunity to succeed. Equal opportunity does not require lower standards.

06 · HEALTHCARE

Healthcare is personal.

Important healthcare decisions belong to people, their families, and their healthcare professionals. Praeva intends to design benefits around meaningful access to legitimate care rather than political fashion.

07 · EXCELLENCE

Kindness and excellence belong together.

A humane workplace is not a low-accountability workplace. We expect people to keep commitments, tell the truth, admit mistakes, protect customers, help colleagues, and care deeply about the quality of their work.

08 · RESPONSIBILITY

Power carries responsibility.

The more authority someone has at Praeva, the greater their obligation to use it well. Nobody gets to hide cruelty, harassment, discrimination, bullying, or abuse behind performance, seniority, revenue, or technical brilliance. That includes leadership.

09 · GROWTH

People can grow.

Accountability matters, but so do education, conversation, forgiveness, and change. Good-faith people willing to listen, learn, and treat others with dignity belong here.

GIVE A DAMN OUTSIDE THE OFFICE, TOO

Praeva matches employee charitable giving.

Praeva will match 100% of eligible employee charitable contributions, up to $2,000 per employee per calendar year. Employees choose the causes, communities, schools, nonprofits, relief organizations, and institutions that matter to them. Praeva helps amplify their impact.

100% match up to $2,000 / year
“Be excellent to each other.”
Bill & Ted's Excellent Adventure

When policies get complicated, our test should stay simple: Are we treating people the way a good company should treat people? If the answer is no, the policy is wrong. Change it.

THE NAME

Praeva means going before.

The name is rooted in the idea of going before or leading the way. That is the role Praeva is built to play: standing before consequential agent actions and verifying the authority behind them before execution.

Before action, authority.

Serious infrastructure. Good humans. No Corporate Oatmeal™.

DESIGN PARTNERS + DEMOS

Build agentic systems that can prove their authority.

Praeva Systems is engaging with security, identity, infrastructure, and AI platform leaders to validate consequential agent workflows.