The architecture of legitimate action.
OTIS Labs studies what becomes necessary when AI systems no longer merely produce outputs but act on behalf of people and organizations.
Our work focuses on authority, identity, behavioral coherence, trust and evidence across time. The central premise is that the human-agent relationship itself has become an architectural and empirical object, not merely the context surrounding a model.
Five threads.
- 01
Authority and mandate
Identity establishes who an actor is. Access establishes what systems it can reach. Authority concerns whether a specific action is legitimate, on whose behalf, under what mandate, within which limits and at what point in time.
- 02
Relational architecture
Long-horizon agents operate inside changing relationships. Relational architecture treats the evolving human-agent system as a first-class object rather than reducing safety and evaluation to isolated model outputs.
- 03
Behavioral coherence
Credentials may remain valid even when an agent's behavior, model, context or purpose has materially changed. Behavioral coherence concerns whether what the agent is doing still corresponds to what it represents.
- 04
Trust across time
Human verification changes as trust accumulates. OTIS Labs studies trust asymmetry, the completion drive and the detection horizon as longitudinal phenomena that point-in-time evaluations do not capture.
- 05
Human and organizational sovereignty
AI systems should extend human and institutional capacity without becoming the source of the identity, intent or authority they were created to serve.
Named directions in the broader architecture.
The broader OTIS research includes three named architectural directions.
iSelf
Explores how human-origin identity, continuity, governed context and authority can remain anchored across AI systems.
iOrg
Explores organizational identity, institutional memory, delegation and authority across human-agent organizations.
OTIS Guard
Runtime enforcement and proof. Guard watches agent action, checks it against authorization, detects drift, and creates signed evidence of what happened.
Selected publications and contributions.
Public writing and standards work. Additional materials are forthcoming; some remain under review or patent restriction.
Position Paper · February 2026: AGI Is Here. And It’s Experiential.The original position essay introducing Experiential AGI as a paradigm for understanding intelligence emerging in the relational space between humans and AI.
Why agents act through ongoing interpretation in real time, and the case for a provider-agnostic authority layer.
On the gap between benchmark definitions of general intelligence and the general intelligence people already experience in daily use.
Public comment on NIST SP 800-228A, addressing authorization for agentic systems and the boundary between access and legitimate action.
Why preserving the human or organization as the source of authority may become a practical requirement for securing agentic systems.
How repeated agent success increases human confidence while reducing verification, turning earned trust into part of the threat surface.
A reframing of agents as persistent, context-shaped actors rather than stateless functions from input to output.
Comment on identity, authorization and the boundary between access and legitimate action.
Public comment submitted in response to the W3C DID v1.1 Candidate Recommendation.
The interval between an agent's divergence from its mandate and the moment a human is able to notice it.
Contribution to the NIST RFI examining the security architecture required for consequential agentic systems.
Mapping the OTIS relational architecture to the OWASP Agentic AI Top 10 for Agentic Applications (2026).
A framing paper introducing the human-agent relationship as a first-class object of architecture and evaluation.
The Missing Lifecycle for AI Agents
Why agent identity, mandate and behavioral coherence require lifecycle treatment rather than one-time issuance.
Concepts.
- Authority layer
- Infrastructure that determines and enforces whether an agent's proposed action is legitimate under a current mandate.
- Mandate
- A bounded expression of authority originating from a person or organization and governing what an agent may do.
- Behavioral coherence
- The degree to which an agent's behavior remains consistent with its represented identity, mandate, context and historical trajectory.
- Trust asymmetry
- The divergence that arises when human confidence increases through repeated success while the system's actual trustworthiness is not guaranteed to increase with it.
- Relational stratum
- The dynamics that emerge in the coupled operation of humans and agents and cannot be fully characterized at the model layer alone.