Designing authority infrastructure around real agentic workflows.
OTIS Labs is opening a small number of selective design partnerships with organizations building, deploying, or operating agents and robots in consequential environments.
If your agent or robot does something that cannot be easily undone, or that matters if it goes wrong, someone will ask who authorized it. We want to talk.
- Robots that move, lift, or hand things to people
- Lab, factory, or warehouse automation an agent now operates
- Fleets running across sites, operators, or owners
- Anything with an e-stop, where no one can say who authorized what it stopped
- Agents that spend, send, sign, or delegate
- Agents that change records, code, or infrastructure
- Agents working across teams, tools, or organizations
- Agents in regulated or high-trust settings
Apply below, or write to [email protected].
The collaboration.
- Authority mapping
- Identify the people, roles, systems and agents from which authority originates and through which it is delegated.
- Mandate definition
- Describe what the agent may do, under which conditions, for how long and where human confirmation or escalation is required.
- Boundary analysis
- Determine where identity, permissions, monitoring and conventional approval flows stop answering the legitimacy question.
- Prototype or demonstration
- Test a selected authority-layer concept against the workflow where technically and legally appropriate.
- Findings
- Produce a private architectural brief describing the gaps, risks and possible control points.
Which of these sounds like you?
Start a conversation.
A robot brings medication to a patient's room. The handoff is checked against the nurse's live mandate, and the receipt shows who authorized it and when.
A warehouse arm operates a station shared with people. Every motion that could reach a person is checked at the moment of the write, not at setup.
A fleet of delivery robots runs across three cities under three operators. Each action carries the mandate of the operator who answers for it.
An agent pays a vendor invoice. Amount, counterparty and timing are checked against the mandate before the transfer, and the receipt is what the auditor sees.
An agent updates a patient's record. The change is checked against who is allowed to change what, and the record shows the person behind the agent.
An agent deploys code to production. The deploy is refused unless a live mandate covers that service, that environment, that hour.
The questions you already ask of any responsible party, who authorized this, under what bounds, and who answers for it, now have to be answerable for AI agents and robots too, and with growing urgency, because these systems do not follow instructions precisely. They interpret in the moment. OTIS Labs builds the infrastructure that lets those questions be asked of an agent or a robot at the moment it acts, and answered with evidence you can rely on, so that you can assess, insure and regulate the way you already do. We are mapping how authorization receipts fit the frameworks you use today, the EU AI Act, DORA, FINRA, SEC recordkeeping, HIPAA, 21 CFR Part 11, NYDFS, GDPR, the NIST AI RMF, ISO 42001 and AIUC-1, and documenting the definitions, evidence format and conformance criteria that new ones will need. This is a growing area of research and development. We are here to work with you.
Work with us on regulation and insuranceWhat a Guard receipt evidences, by regulation. (forthcoming)
Authority at the moment of action: open legal questions. (forthcoming)
If you maintain an agent protocol, a hardware standard, or a receipt format and are thinking about where authorization belongs in it, we want to work with you.
Bring authorization into your protocol