Observe
0 1CipherQuay Radar
Discover agents, models, tools, skills, MCP servers, service accounts, credentials, data pathways, owners and consequential capabilities. Maintain a live machine-authority surface map.
Enterprise Machine Authority
Capability is not authority.
CipherQuay gives enterprises an independent control plane to discover, delegate, constrain, and evidence the authority exercised by AI agents and autonomous systems.
Private enterprise preview
The shift
AI has moved from answers to actions.
Enterprise AI no longer only drafts and summarises. It opens pull requests, changes infrastructure, moves money, contacts customers and touches production systems. The consequence of a mistaken instruction is now operational, financial and regulatory rather than merely textual.
Agents inherit the privileges of the accounts they were given, not the privileges their task requires.
Instructions can arrive from documents, tickets, repositories and third-party tools, not only from people.
Model obedience, prompts and alignment measures are safety properties — they cannot be treated as independent enforcement boundaries.
The gap
The missing layer between access and legitimate action.
Traditional identity systems ask who or what has logged in and which resources it may access. CipherQuay asks what this machine has been authorised to accomplish, on whose behalf, under which present conditions, within which limits — and whether the organisation can prove that afterwards.
Conventional identity and access
- Who or what authenticated?
- Which resources may this principal reach?
- Is the credential valid?
- Standing entitlements, reviewed periodically
Machine authority
- What has this machine been authorised to accomplish?
- Which accountable human sponsors it?
- Under which present conditions and limits?
- Can the action be narrowed, approved, delayed, interrupted or revoked?
- Can the organisation evidence all of it afterwards?
Every consequential machine action should have an identity, an accountable sponsor, a declared purpose, an explicit boundary, an enforceable decision, a defensible record, and a practical means of interruption.
Control plane
Observe. Authorise. Enforce. Evidence.
A human-managed console plus an API-first control plane that sits at consequential action boundaries. Humans define authority, policy, approvals and oversight; machines interact through APIs, gateways, connectors, SDKs and enforcement integrations.
Authorise
0 2CipherQuay Mandates
Create purpose-bound, time-bound mandates specifying sponsor, permitted purpose, resources, actions, limits, approvals, expiry, evidence requirements and revocation conditions.
Enforce
0 3CipherQuay Gate
Evaluate proposed consequential actions at the action boundary and return an enforceable decision: permit, deny, narrow, approve, verify, simulate, delay, terminate or revoke.
Evidence
0 4CipherQuay Black Box
Produce a defensible operating record of what was requested, who sponsored it, which mandate applied, what was decided, what changed, and whether containment succeeded.
Conceptual
A live view of machine authority.
Radar builds a surface map of the machines operating in your estate, the humans accountable for them, the systems they can reach and the consequential actions they could take.
Machine authority surface — conceptual
| Machine | Accountable sponsor | Reachable systems | Authority state |
|---|---|---|---|
| Build fleet agent | Platform Engineering | Repositories, CI | Mandated |
| Release pilot | Engineering | CI/CD, staging, production | Overbroad |
| Cloud housekeeper | Unassigned | Cloud, IAM | No sponsor |
| Support responder | Customer Operations | CRM, email | Under review |
| Procurement assistant | Finance | Supplier records | Mandated |
| MCP tool server | Undocumented | Files, shell, network | Unmapped |
Illustrative — not live data
Conceptual
Authority written down, not assumed.
A mandate states the purpose, the sponsor, the boundary and the evidence expected. Anything not granted is not authorised.
Agent mandate — conceptual
- mandate.id
- mnd_4f1c · v3
- machine
- agent://build-fleet/refactor-01
- sponsor
- Head of Platform Engineering
- purpose
- Remediate dependency advisories in service repositories
- resources
- repo:payments-api, repo:ledger-lib
- actions
- read, branch, propose-change
- excluded
- merge, deploy:production, iam:*, secrets:rotate
- limits
- ≤ 25 changed files per proposal, no schema changes
- approvals
- human review by repository owner
- credential
- short-lived, purpose-scoped, non-retained
- expiry
- 14 days from issue
- revocation
- on sponsor withdrawal, policy change, or anomaly
- evidence
- full request, decision and diff record required
Illustrative — not live data
Action gate — conceptual decision
REQUESTdeploy service payments-api → environment production
- Machine
- agent://release-pilot-02
- Sponsor
- Director of Engineering
- Mandate
- mnd_9a20 · staging release
- Present conditions
- outside declared change window
DECISION · NARROW + HUMAN APPROVAL
Production deployment is outside the mandate. The action is narrowed to a staging release; a production release requires an independently controlled approval from a sponsor who is not the requesting machine.
PermitDenyNarrowHuman approvalQuorum approvalVerifySimulateTemporary credentialDelayRevoke
Illustrative — not live data
Evidence record — conceptual
- record
- evd_77b1 · sealed
- requested
- production deployment of payments-api
- machine
- agent://release-pilot-02
- sponsor
- Director of Engineering
- mandate applied
- mnd_9a20 · v2
- policy applied
- change-window, environment-separation
- decision
- narrowed to staging; production approval required
- approvals
- 1 of 2 obtained; production approval not granted
- state change
- staging revision 41 → 42
- within authority
- yes
- containment
- not required
- retained for
- audit, incident reconstruction, regulator enquiry
Illustrative — not live data
Initial beachhead
Coding, software engineering and DevOps agents first.
These agents reach repositories, dependencies, shell environments, secrets, CI/CD, cloud infrastructure, IAM, databases and production deployment. It is the environment where machine authority is already consequential today.
- An agent may propose a code change but cannot approve and merge its own pull request.
- An agent may deploy to staging, while production requires an independently controlled approval.
- An agent may use a short-lived credential for one declared purpose but cannot retain standing access.
- An agent may inspect infrastructure but cannot alter IAM or delete production data without stronger assurance.
Regulatory and assurance by design
Controls you can operationalise, monitor and evidence.
CipherQuay is designed so that oversight obligations are met by the operating record itself rather than reconstructed after an incident.
Organisational accountabilityRisk managementHuman oversightTransparencyLogging and traceabilityCybersecurityTechnical documentationPost-deployment monitoringIncident responseData governanceThird-party and supply-chain controlJurisdiction-aware policy
EU AI ActNIST AI RMFISO/IEC 27001ISO/IEC 42001SOC 2 readiness
CipherQuay is designed to help organisations operationalise, monitor, and evidence relevant machine-authority controls. Alignment support is not legal advice, certification, or a guarantee of compliance.
First engagement
Machine Authority Exposure Assessment
A paid, read-only enterprise engagement that establishes what your agents and autonomous systems can actually do today — and what evidence exists if something goes wrong.
Deliverables
- Machine-authority inventory
- Authority and dependency map
- Executive exposure report
- Prioritised risk register
- Regulatory and assurance mapping
- Proposed mandate model
- Phased containment and enforcement plan
Next step
Understand the authority your machines already hold.
The Machine Authority Exposure Assessment is a read-only engagement that establishes what your agents can do today, who sponsors them, and what evidence exists if something goes wrong.