Bounded autonomy
Every agent operates inside a written remit, enforced by the kernel at runtime. An instruction outside the remit is refused and logged.
Governance, security & ethics
Firms do not hesitate over AI because the technology is weak. They hesitate because nobody can tell them, precisely, what it will and will not do. This page is our answer, in full, published where your risk team can read it before we ever meet. It covers three distinct things: how agents are governed, how your data is secured, and where we stand on the responsible use of AI.
-
SOC 2 · ISO 27001 · GDPR
Aligned with SOC 2, ISO 27001 and GDPR
-
Australian Privacy Principles
Aligned with the Australian Privacy Principles
- Ask for the current evidence pack at any briefing.
Governance
Oversight, accountability and control: who decides, who approves, and how the platform enforces it at runtime.
Sign-off gates
Consequential actions stop and wait for a named approver. The gate schedule is agreed with you at deployment and changing it is itself a gated action.
Append-only audit trail
Every action, draft, approval and refusal is written to a tamper-evident log. No one can edit or delete an entry, including us.
Stand-down control
Any authorised person can stand down an agent, or the whole roster, immediately. Work in flight is parked safely and the stand-down is logged.
Security & privacy
Data protection, infrastructure and access: how your data is stored, moved, encrypted and reached.
Data handling by agreement
Where customer data is stored and processed is set out in your agreement. Processing runs under data processing agreements and retains nothing.
Role-based access
Agents hold the minimum access their role requires, granted through your identity provider, reviewed on a schedule, and revocable in one action.
Encryption throughout
Data is encrypted in transit and at rest. Keys are managed per customer, and key operations appear in the same audit trail as everything else.
No training on your data
Your data is used to do your work, and for nothing else. It is never used to train models, ours or anyone else’s, and this is a contractual term.
Built around what you answer for
Your risk is not abstract. It has section numbers, and a person whose name is on it. These are the obligations we designed the controls around, international standards first, with regional frameworks inside that picture.
GDPR · Art. 5 & 32
Personal data, processed lawfully and securely
For European deployments, and for any firm serving people in Europe: processing runs under data processing agreements at every layer, access is least-privilege and logged, and your data is never used to train a model. Where data is stored and processed is written into your agreement.
Privacy Act 1988 · APP 11
Security of personal information (Australia)
You must take reasonable steps to protect the personal information you hold. Agents work inside your identity provider with least-privilege access, every access is logged, and cross-border processing is disclosed and covered by data processing agreements rather than buried.
Professional codes · APES 110 and equivalents
Codes of ethics for professional advisers
Confidentiality, professional competence and due care do not delegate to software, under any professional code. Agents draft; a named person reviews and signs. The decision log records who approved what, and on what information, so the professional judgement stays evidenced and yours.
Operational resilience · DORA · CPS 234
Information security and operational risk
If you are regulated for operational resilience, in any jurisdiction, the questions are about material service providers, incident notification and tested controls. Hosting, access, logging and continuity terms are written into the agreement, where your auditor can read them.
Security baselines · NIST CSF · Essential Eight
The baseline your clients ask about
Application control, patching, hardening, backups, access restriction, multi-factor. We track our maturity against the baselines your jurisdiction uses and share where we sit at any briefing.
Record-keeping · IPPF
Records that stand up years later
Statutory records must be complete, retained and reproducible on request, and internal audit needs the evidence itself. The audit trail is append-only and hash-chained, retained on your schedule, and exportable in standard formats, so fieldwork years from now examines the record directly.
Not legal advice, and not a claim that software discharges your obligations. These are the standards we build to so that meeting them is evidenced rather than asserted.
Ethics: built to augment your people
Legare Global is built to accelerate the people a firm already employs. Agents automate the mundane, repetitive and administrative work, and the time that frees flows to strategy, decision-making, management and the other work that needs a person.
Augmentation
Employees stay at the centre
Legare Global agents are designed to accelerate and augment the output of the people you already employ. They take on the mundane, repetitive and administrative work, and the role a person holds stays theirs.
Higher-value work
Time moves up the value chain
The hours an agent absorbs come back as time for strategy, decision-making, management and client relationships: the work that needs a person. The measure of a deployment is what your team can now spend its attention on.
Human oversight
A person stays in charge
Every consequential action stops at a named approver, and every action is logged. Oversight is engineered into the platform, so responsible use holds at any scale.
This position will keep developing as standards for responsible AI mature. As it does, it will be published here, in full, the way the rest of this page is.
Governance in depth
01 — Bounded autonomy
The remit is a technical control
Most AI governance fails at the same point: the rules live in a document while the system does whatever it is asked. On the kernel, the position description you approve is compiled into runtime constraints. The Accounts Receivable Controller cannot alter an invoice because the capability is absent from its remit. When an agent is instructed to act outside its remit, by anyone, it refuses, and the refusal is logged with the instruction that triggered it.
02 — Audit trail
Built for the person who has to answer for it
The audit trail is designed for your risk committee, your external auditors, and the worst day you hope never comes. Every entry records the actor, the action, the inputs relied on, the output produced, and any human approval attached. Entries are hash-chained so tampering is evident, retained on your schedule, and exportable in standard formats. When your auditor asks how a decision was made, the record answers in minutes.
03 — Where data lives
Disclosed in your agreement.
Where your data is stored and where it is processed are written into the agreement, not the FAQ, and you can verify both: we will show you where your data lives. Processing, including model inference, runs on secured infrastructure; every hop is covered by a data processing agreement, nothing is retained, and nothing is ever used to train a model. Remote support access is disabled by default and, if you enable it, appears in the audit trail like any other access.
04 — Access controls
Least privilege, for agents and for us
Agents authenticate through your identity provider and hold the minimum access their role requires. Access reviews run on a schedule you set. Our own staff have no standing access to your data; any support access is requested, time-boxed, approved by you, and logged. Offboarding an agent, or the whole platform, revokes access in one action and leaves you with a complete export of your data and your audit trail.
Book a demo
Tell us the role you would otherwise hire for. We will show you an agent at work on it in a sample firm: hand it a task and watch the sign-off gate hold its first draft. No deck, no obligation.