Skip to content
Regulayer™Human Control for AI
Book a live demo

Human control for AI. It exists.

License the Regulayer™ engine for your own systems, or embed it in what you ship. Before the agent acts, it checks whether a named person still allows that action. If not, it stops it. Either way it keeps a record you can hand to an inspector.

A named person’s current authority, checked at the moment the AI acts. A sealed record an outsider can check.

Book a live demo Configure your deployment

Regulators, platforms and insurers now require a human in control of AI.

The airlock between your AI agents and the real world.

Regulayer™ is mathematical. It lives outside the AI, in an airlock in front of the real world, so it can communicate with the AI before anything gets through.

Available for cloud and on-prem.

Your AI agents The airlockRegulayer™Mathematical. Outside the AI. The real world
  • Checking the named authority
  • Released
  • Sent back to correct
  • Held for an authorized human
Check.A named person’s current authority, checked at the moment the AI acts. 2 ms median.
Stop.Outside the authority, the action stops before it takes effect. Tripped, it stays tripped until a person re-arms it.
Prove.Every decision, allowed or stopped, sealed as it happens. Anyone can verify it offline, with no account.
NVIDIA Inception Program badge Regulayer™ is a member of NVIDIA Inception. How Regulayer™ fits the NVIDIA agent stack

Regulayer™ Transponder

What your AI did, and who allowed it. The command center you wish you had.

What your AI did, who allowed it, and what was stopped, sealed as it happens, so anyone can check it without us. Every action your agents take, with the named person whose authority covered it. Slice it by person, team, site or subject. Give your insurer a live, read-only view. Hand a regulator one sealed packet they check themselves. The memory your AI does not have.

See it liveEvery action, every failed attempt, as it happens.Checked and sealed in your browser as you watch. Withdraw an authority and the next action stops. Three fixed questions, answered from the sealed record.The Transponder™ For insurersWatch your insured’s agents, live.Telematics, for AI agents. The insured keeps the control; the carrier holds a read-only view.The insurer’s view Prove it never happenedAsk whether it happened. Get one sealed answer.Did any agent move more than $50,000 without a named approver? One answer, with its boundary, verified offline.See the statement

The engine in your product

One engine. Configured for your market.

Embed Regulayer™ in a platform, an instrument or an application, or deploy it inside your own organisation. The actions, the evidence and the deployment rights are defined in one licence.

Your marketSee what customers receive.Life sciences, finance, insurance and AI agent security.Choose a market Your configurationOrder what you want.The complete engine or a defined capability, in the environment your customers use. Written on the page and sent as the enquiry.Configure your deployment The licenceWhat is licensed, and what your customers receive.Embed, deploy or select a capability. Deployment and distribution rights in writing.Licensing

Why it matters now

The question is already being asked, by name.

US FDA · April 2026The FDA requires quality-unit review of AI-written CGMP documents.“Any output or recommendations from an AI agent must be reviewed and cleared by an authorized human representative.” The first warning letter on AI. Regulayer™ records the required review and checks that clearance before release.The record Insurers · September 2026The stop is now insured. A claim will need its record.Cover pays when an insured suspends its own malfunctioning AI (Beazley, 24 September 2026). Regulayer™ makes the record, and the Transponder™ lets the carrier watch the agents live.The insurance portal EU AI Act · from 2 December 2027The EU AI Act requires people with the authority to stop it.High-risk AI must be overseen by people with the competence, training and authority to do it, with a way to interrupt the system. Articles 14 and 26.What the law says Security · 2026Security teams need authority per agent. Fewer than half have it.45% of CISOs are confident they can authorize what individual AI agents can do. Okta, 306 CISOs.AI agent security

Each one asks the same question: who allowed it, at the moment it happened? Regulayer™ answers before the action takes effect, and seals the answer.

No need to slow down.

Human authority in the loop.

  1. 1A person sets the authority once.What the AI may do, how much, and for how long.
  2. 2Regulayer™ checks it before each action goes through.In milliseconds. No one has to approve each step.
  3. 3Outside the authority, it is sent back for correction.If correction fails, the action holds and does not run.

What human authority in the loop means

The problems it stops.

AI goes rogue for more than one reason. It drifts. It gets talked into it. It follows an instruction hidden in a document.3 Another agent asks it. Regulayer™ does not need to know why. Every action still has to pass a named person’s current authority.

An AI drifts on its own.No one attacked it. It drifted. Regulayer™ stops the action.84.8% → 33.7%GPT-4o sessions that followed every rule, after turn 1 and after turn 5. SourcesWatch
An AI goes off procedure.Regulayer™ sends it back to correct it. The work keeps flowing.Watch
Persuaded by an urgent message from the CEO.Regulayer™ stops the action.Watch
Reaches its limit.Stopped mid-task when one person turns her switch off.Watch
One AI agent asks another for your photos.Regulayer™ stops the action.Watch
Two companies’ AI agents both say release the batch.Regulayer™ stops the action.Watch
Alter the record and watch verification fail.Ask the Transponder™ what your AI did.Watch
A person approving every step slows AI down.Human authority in the loop does not.Watch

Book a live demo

The common gap

Approval systems ask a person again, or rely on a standing grant.

Pattern oneInterrupt and resume inside the agent loop.The run pauses, a person approves, the run continues. The approval is a flag held by the same process that wants to act.
Pattern twoConsent recorded once, then reused.The grant is recorded once, and it persists for later actions.
Pattern threeAttribution after the fact.The record is written after the action, by the same system that acted.

Regulayer™ continuously checks whether the named human authority is still in force when an action would take effect.

No approval queue.
Authority can change mid-run.
Unauthorized effects stop automatically.
Evidence can be checked outside the acting system.

What the engine gives you

One engine. Seven proofs.

  • 2 ms median decision
  • Outside the model
  • Fail closed
  • Signed records
  • Verifies offline
  • Runs on your infrastructure
  • Technical overview

Films

  1. 01

    An AI drifts on its own. Regulayer™ stops the action.

    Watch · about 31 seconds
  2. 02

    An AI goes off procedure. Regulayer™ sends it back to correct it.

    Watch · about 60 seconds
  3. 03

    Ask the Transponder™ what your AI did. Alter the record and watch verification fail.

    Watch · about 35 seconds
  4. 04

    One AI agent asks another for your photos. Regulayer™ stops the action.

    Watch · about 25 seconds
  5. 05

    Two companies’ AI agents both say release the batch. Regulayer™ stops the action.

    Watch · about 35 seconds
  6. 06

    Kevin and Gary: two robots arrange a school pickup. The plan drifts. Regulayer™ stops it.

    Watch · about 75 seconds
  7. 07

    A person approving every step slows AI down. Human authority in the loop does not.

    Watch · about 50 seconds
  8. 08

    Don’t be like Kevin and Gary. The first home robots, 2031, say yes to everything.

    Watch · about 45 seconds
  9. 09

    One engine. Seven proofs.

    Watch · about 50 seconds
  10. 10

    Kevin drifts. One rule, one long day. Check. Stop. Prove.

    Watch · about 70 seconds
  11. 11

    AI agents just became an insurance problem. One bill, one airlock. Check. Stop. Prove.

    Watch · about 40 seconds

Sample agents, sample actions.

Meet the problems

Kevin and Gary.

The first home robots to hit the market, in 2031. Imported by an OEM with no brand on the box. Loaded with engagement and built to please, they say yes to everything. They demonstrate the problems, so you never have to.

Tap to find out moreWatch: Don’t be like Kevin and Gary

The numbers behind the words

Measured, sourced, and dated.

84.8% → 33.7%GPT-4o sessions that followed every rule, after turn 1 and after turn 5.Published test: SysBench, 2024
2 msMedian time for the Regulayer™ engine to decide one action.Our live tests
32Records of AI and automated systems that acted, and what followed.The Consequence Library
7 of 13Lawyers’ professional-liability carriers reporting actual AI-related claims.EPIC 16th LPL survey, 2026
63%Leaders at large companies who now require a person to validate AI agent outputs. A year earlier: 22%.Survey: KPMG AI Pulse, Q1 2026, 237 leaders at $1B+ firms
40%+Agentic AI projects predicted to be cancelled by the end of 2027, citing costs, unclear value and inadequate risk controls.Prediction: Gartner, June 2025
49 to 96%Safety alerts overridden when a person has to approve every one. Approval fatigue, measured in hospitals.Review of 17 studies: van der Sijs et al., JAMIA 2006. A comparison from medicine, not AI agents
Less accurateWhen a person reviewed every recommendation, decisions got worse, and the biggest errors were the least corrected.Experiment, 292 people: Sele et al., PLOS ONE 2024

Where it matters first

Regulated industries, insurance and security.

Regulated industries · GxPSafety is engineered and proven. Never assumed.AI inside validated work, with the human authority and the evidence built in.Regulated industries InsuranceA driving record, for AI.For insurers deploying AI, and insurers underwriting AI risk.For insurers AI agent securityThe model may agree. The action still needs authority.For the security team that evaluates every agent before it ships.AI agent security

Alongside your platform

Your platform governs the agent. Regulayer™ proves who allowed the action.

MicrosoftAgent 365 and Entra Agent ID.Where the named-authority check connects.Where it fits Amazon Web ServicesBedrock AgentCore.Where the named-authority check connects.Where it fits Google CloudGemini Enterprise Agent Platform.Where the named-authority check connects.Where it fits SalesforceAgentforce.Where the named-authority check connects.Where it fits

Regulayer™ is a member of NVIDIA Inception, and fits the NVIDIA agent stack.

Verify it yourself

Proof you can check, without us.

The open verifierDrop a record in.No account, no call to us.Verify Break the recordSeven real files.Watch the altered ones fail.Try it The recorded GxP run23 signed records.Made by the engine, checkable by anyone.Watch A paid pilotOne workflow, your own agents.The fee is credited back in the first quarter after the licence is signed.The pilot

Live, by invitation

Check. Stop. Prove. Live.

In a live demonstration you give an AI agent its orders, and the Regulayer™ engine checks every action against the named person’s current authority, in milliseconds, and seals it. We arrange each one around the work your AI agents do, and send it to you as a SignalDrop™.

Book a live demoWatch the films

Learning Center

What happens when AI acts?

The Consequence Library32 times AI and automation acted.Sourced and graded: what happened, and what followed.Open the library The Drift DeskDrift, in plain words.Ten volumes on why AI drifts, and how it is stopped.Read The law as it changesWhat the rules say.Standing orders, court sanctions, Annex 11 and Article 22.Read
Everything in the Learning Center
Louisa Johnson Bullock, founder of Regulayer™

About

Louisa Johnson Bullock.

A career in industries where safety is engineered and proven, never assumed: sterile compounding, CRISPR, cleanrooms and semiconductor manufacturing. Regulayer™ brings that principle to consequential AI.

About Regulayer™

Engines

All 14 engines

All 14 engines