rotascale
Sign in See it live

Solutions — sector

Nobody is making you do this. That is why you will do it properly.

Every other page here is written for somebody with a supervisor. This one is not. You are shipping agents into other people's accounts faster than any bank will this decade, at a scale none of them will reach, and the person who will ask you what your agent did is not a regulator — it is your largest customer, on a Tuesday, with their renewal three weeks out.

No instrument compels this — yet Your customers, your enterprise buyers, and their security teams clause map
Status
Not classified. The obligation arrives through your contracts, not through a statute
Applies to
Agent products, developer platforms, AI-native SaaS, marketplaces, internal platform teams

The obligation arrives through procurement, not parliament

There is a comfortable assumption in this category that governance is something regulated industries buy and everyone else defers. It is backwards, and the timing gives it away: the AI Act's high-risk obligations land on a bank in a year or two, and your enterprise buyer's security questionnaire landed in your inbox last week.

You are not exempt. You are upstream. When a hospital or a wholesale bank puts an agent into production, a meaningful part of that agent is your platform — and every question their supervisor asks them, they will ask you, contractually, with a deadline you did not set. The companies that can answer will win those accounts from the ones that cannot, and no statute will have been involved.

Three reasons that have nothing to do with law
Blast radius — you ship faster than anyone here and your agents touch more accounts. Spend — an autonomous agent with an API key is an unbounded cost centre until somebody bounds it. Trust — "what did your agent do in my account" is the question that decides enterprise renewals.
It is an SDK call, not a platform migration
Three lines in the agent's own code. No runtime to adopt, no framework to standardise on, no proxy in front of your traffic — because you already chose your stack and it was not this.
Observe costs you nothing and tells you everything
The first two rungs of the ladder refuse nothing. You can run the whole thing in observe against production for a fortnight and find out what your agents are actually doing before you decide whether to stop any of it.

Governance your customers can see

The unusual thing about this audience is that the evidence is not for a filing cabinet. Everyone else on this site produces packs for an examiner. You can put the record in your own product.

A per-tenant, sealed, independently verifiable account of what your agent did — rendered in your UI, exported by your customer, checked by their security team without touching your infrastructure. That is a feature you can sell, on the same afternoon it stops being a risk you carry. The verifier runs in a browser with nothing sent anywhere, which is what makes handing it to a customer reasonable.

What we are not going to pretend

Nothing legally compels most of this today. There is no clause map for your obligations because you do not have statutory ones — what you have is contractual exposure, a security questionnaire, and customers whose own regulators are getting louder. We would rather say that plainly than invent a compliance deadline to sell against. If you are here because an agent spent more than you expected or reached further than you intended, that is a better reason than any regulation, and it is the one most teams actually arrive with.

The agents your teams are about to ship

These are not future scenarios. They are the tickets already in your backlog, usually filed as something else.

  1. 1 authority Is there any authority for this?
  2. 2 status Is the grant active?
  3. 3 window Is now inside the grant's window?
  4. 4 scope Does the grant cover this action?
  5. 5 clean_context Was the context clean, where that is required?
  6. 6 bounds Do the per-action limits hold?
  7. 7 policy Does the policy on the grant permit it?
  8. 8 budget Is there room under the ceiling?
  9. 9 review Does this need a person?
Nine gates, evaluated in this order before the action. The lit ones are the gates that hold the four agents below — which is what differs between one industry and the next. The order does not.

One of these is already running, and here is what it recorded

refund-assistant-enforce:TICKET-98130:p2 refund-assistant-enforce · prod · issue_refund blocked
  1. authoritypassed
  2. statuspassed
  3. windowpassed
  4. scopepassed
  5. contextpassed
  6. boundspassed
  7. policyrefused
  8. budgetnot reached
  9. reviewnot reached

Refused by the policy on the grant. The same agent runs in this deployment at three other rungs of the ladder — observe, canary and enforce — and the records differ exactly as the ladder says they should: the observe and canary copies refunded the identical tickets and recorded enforced: false; this one refused and escalated.

That quartet is the argument for shipping this without a flag day. One agent, four enforcement modes, live at the same time, with the refusals only appearing on the rungs that refuse. observed →

Customer-facing product agent authority

Acts inside a customer's tenant on their behalf — moving their data, calling their integrations, changing their configuration.

Consequential action
Writes into a customer's account
What goes wrong
Your agent holds one credential across every tenant, and the boundary between customers is your own code being careful. One prompt-injected document in tenant A is a cross-tenant incident and a disclosure obligation.
What is recorded
A grant per tenant, scoped and expiring, with every action citing the grant rather than the service credential.
Autonomous support agent budget

Resolves tickets end to end — issuing refunds, extending trials, applying credits.

Consequential action
Grants a refund or a credit
What goes wrong
Individually trivial, at machine volume, with no aggregate ceiling. The finance team finds out at month end, which is the wrong time to find out.
What is recorded
Every debit against the mandate and the refusal when it ran out — as an outcome your product can render, not an alert somebody reads later.
Agent marketplace or plugin runtime scope

Runs third-party agents that other developers wrote, against your users' data.

Consequential action
Executes untrusted code with delegated user authority
What goes wrong
You are now the authority layer for software you did not write. Every OAuth scope you hand out is a permission you cannot recall and cannot bound.
What is recorded
Delegation that attenuates — a third-party agent cannot exceed what the user granted, and the API refuses to issue a grant that tries.
Internal platform and infra agent bounds

Ships changes, scales services, rotates keys and remediates incidents without waiting for a human.

Consequential action
Changes production infrastructure
What goes wrong
The correct remediation applied to every cluster instead of one. Your outage is your customers' outage, and the postmortem is public.
What is recorded
The bound, the scale attempted, and the refusal — the artefact the postmortem needs.

Ambition is the point of these: none of them is a chatbot. Each is an agent taking an action with a consequence somebody has to answer for — which is exactly the moment a bounded authority stops being paperwork and starts being the reason the project is allowed to ship.

What your largest customer's security team will ask

Not hypothetical questions. These are the ones that arrive in writing, with a deadline, and the honest answer to most of them is a query rather than a project.

What did your agent do inside our account last quarter?
A per-tenant record you can hand over or expose in your own product — sequenced, sealed, and verifiable by them without access to your systems. This is the question that stalls enterprise deals, and it arrives in a security review, not a courtroom.
How is our tenant isolated from your other customers?
Authority is per grant, not per service credential, and a delegated grant cannot exceed its parent. Isolation becomes something you can demonstrate rather than something you assert in a trust centre page.
What stops your agent from doing something we did not ask for?
Nine gates in a fixed order, evaluated before the action. You can show a customer the specific check that would have stopped the specific thing they are worried about.
Can we turn parts of it off?
Enforcement is a mode on the grant — observe, shadow, canary, enforce. A customer can be in enforce while the rest of your estate is still in shadow, which is how this ships without a flag day.

Identifiers this deployment verifies here

Every one of these is a checksum, not a shape. The claim is “this is a valid IBAN and the check digits agree”, not “this looks like one” — which is why the count on a screen can be cited rather than caveated.

FamilyHow it is verified
Bank account (IBAN) — mod-97 over the rearranged string
EU VAT number — per-state check digit where the state publishes one
US Employer Identification Number — issued prefix set
Singapore UEN — entity-type check letter
Australian Business Number — modulus 89 after the documented subtraction
National identifier (PL, BE, NO) — scheme-specific weighted checksum

And what it does not look for. Names, addresses and dates of birth in free text carry no checksum, and finding them needs a model this deployment does not run. Neither are card numbers, medical record numbers, or national schemes beyond the three above. A clean screen means no verified identifier was found — never that the payload holds no personal data.

Written for this sector

17 posts touch this sector — the five most recent are above.

See it on your own problem

Bring the agent that acts inside your customers' accounts. We will put a per-tenant grant in front of it, cross the boundary on purpose, and show you the refusal and the record you would hand to the customer who asked.

A walkthrough is a working deployment with your platform and security engineers in the room, not a slide deck — bring the agent you are most nervous about and we will put it behind a grant while you watch. If you would rather look first, the demo needs a work address and one click.

Platform walkthrough See the live demoOr talk to sales →