# Building the human-authority layer for autonomous intelligence.

As agentic AI moves from answering questions to acting on behalf of people and organizations, identity and access are no longer enough.

OTIS Labs studies and builds the architectures needed to keep authority human-originated, bounded, revocable and provable at runtime, even as AI agents and autonomous systems operate across systems and over time.

- Explore the research: /research
- View the runtime mandate demonstration: /work/runtime-mandate-check

## In time

- Principal: the source
- Mandate: bounded authority
- Agent: the runtime interpreter
- Action: checked in the moment
- Receipt: verifiable evidence

Human intent becomes a bounded mandate, interpreted by the agent at runtime, checked at the moment of action, proven by a receipt.

## The thesis

Intelligence is becoming abundant. Authority is not.

An agent can hold valid credentials, use approved tools and still act outside the present intent of the person or organization it represents.

The authority layer addresses the questions that identity and access alone cannot:

- Who does this agent represent?
- What mandate is it operating under?
- Is that authority still live?
- Is this action legitimate now?
- Can the basis for the decision be proven?

OTIS Labs is researching the infrastructure required to answer those questions across human-agent and organizational systems.

## Selected work

1. Working demonstration, MIT Project NANDA, 2026: OTIS Guard: Runtime Mandate Check. An AI agent receives bounded authority to spend. Each proposed action is checked against a live human mandate at the moment of action and returns approve, confirm, escalate or block. View the demonstration: /work/runtime-mandate-check
2. Position paper, SSRN, July 2026: Experiential AGI Is Already Here. Intelligence, alignment, and human sovereignty in the human-AI relationship: the position paper behind the OTIS research program. Read the position paper: /research/experiential-agi-is-already-here
3. Public standards contribution, W3C: Behavioral Coherence in Agent Identity. A public comment examining how an agent can retain valid identity and credentials while the behavior or cognitive substrate behind that identity has materially changed. Read the contribution: /research/behavioral-coherence-in-agent-identity-subjects
4. Working research: Trust Asymmetry and the Completion Drive. An examination of how repeated agent success increases human trust while reducing verification, turning earned confidence into part of the threat surface. Read the research: /research/trust-asymmetry

## Current directions

- Authority and mandates: how legitimate authority originates, is delegated, changes, expires and is evidenced when an agent acts.
- Behavioral coherence: how to determine whether agent behavior remains consistent with its identity, mandate, context and history over time.
- Relational architecture: how the evolving human-agent relationship can become a first-class object of architecture and evaluation.
- Human and organizational sovereignty: how people and institutions retain control of identity, memory, intent and decision-making as intelligence becomes increasingly delegated.

## Design partnerships

Consequential agents require real authority models.

OTIS Labs is opening a small number of design partnerships with organizations preparing to deploy agents into consequential workflows.

We are particularly interested in robots that move, lift or hand things to people, fleets running across sites or owners, and agents that can spend or move money, send external communications, deploy code, modify sensitive records, approve decisions or act across organizational boundaries.

Explore a design partnership: /design-partners

## Stay in touch

Updates from the lab. Occasional notes on research releases, product launches and public contributions. Subscribe at https://otislabs.ai/ or write to hello@otislabs.ai