Governed Autonomy
Research question
How far can machine work be delegated before the delegation itself becomes the risk?
Working thesis
Autonomy should expand only where evidence supports it. Capability envelopes, staged authority, explicit approval gates, revocation, and recovery are the mechanisms that make delegated execution bounded — and constitutional constraints belong in the runtime, not in a document nobody enforces.
A thesis is the lab’s current position, not a result. What supports it is listed below; what remains unresolved is stated at the end of this page.
Publications
5 attributedThe TripSitter's Guide to AI Hallucinations
Operational Reliability Research
Research ReleasePublishedJanuary 3, 2026Governed AI Systems for Real-World Operations
The shift from isolated model responses to governed execution.
Technical NotePublishedDecember 3, 2025Staged Autonomy and Human Review
Autonomy should expand only where the evidence supports it.
Research DirectionPublishedSeptember 17, 2025Defensive Runtime Research Without Exploit Publication
Studying agent failure without arming attackers.
Research DirectionPublishedAugust 3, 2025The Eve Constitution as Runtime Governance
A public governance charter for agentic systems and AI workspaces.
Constitutional ArtifactPublic DraftJuly 17, 2025
Apparatus
5 systemsSystems attached to this program are what the lab builds to investigate the question. Maturity, evidence level, and current activity are recorded separately on each system — a system attached here is not thereby claimed to be production-ready.
EXO
AI Workspace / Operator Product
A public AI-native desktop workspace for sustained work across models, agents, code, tools, projects, tasks, context, approvals, and connectors.
Cerberus
Multi-Head Reasoning & Evaluation Harness
A three-head reasoning and evaluation harness built on structured disagreement: one head proposes, one head critiques adversarially, and one head arbitrates and verifies.
TAG
Typed Agent Gateway
A typed delegation protocol that models every handoff between agents, tools, models, and humans as a structured object with explicit authority, scope, budgets, and evidence contracts.
TripSitter
Supervisory Runtime
A supervisory runtime for active monitoring of autonomous AI sessions: uncertainty detection, drift detection, budget and policy monitoring, and human escalation with bounded intervention.
AMR
Agent-Machine Runtime Research
A research program for connecting AI agents to controlled machine interfaces while preserving observability, permission boundaries, and recoverable execution.
Open questions
Published as open. These are the parts of the program the lab does not consider settled.
- 01
Which evidence actually justifies widening an envelope, and which merely looks like it does?
- 02
How is oversight kept meaningful rather than ceremonial as volume grows?
- 03
What is the right response to a system that is confidently wrong rather than visibly broken?