Security and compliance
Your clients' secrets, kept by design.
Confidentiality is a duty you cannot hand to a vendor, so we are building a platform that lets you keep it: your own database, a key for your firm, a design for Australian residency or nothing leaving your server at all, and a security monitor that watches the whole platform and never strikes back.
Each control below says whether it is built and in alpha testing on our test server, in development, or planned. We hold no certification and claim none. What is still open is listed at the end, with its milestone.
Who it protects
Security that answers the question each reader is asked.
-
For law firms
Your duty of confidentiality does not stop at your own network.
- Your own database, and client text sealed with your firm's own key In development
- Encrypted LexIQ backups with no unencrypted copy left on disk Built, in alpha testing
- Threat notifications and a monthly security report for your firm In development
- A second sign-in step, always required for administrators Built, in alpha testing
What is still open is listed at the end of this page with its milestone.
-
For in-house counsel
Boards ask what could go wrong and how you would know.
- A monthly threat report scoped to your organisation In development
- Defensive-only security monitoring that never strikes back In development
- Export and erasure on request, with an audit trail Built, in alpha testing
AegisIQ's detections are built and measured against replayed attacks, and run live once the rule set is signed off; alerts are in progress and not yet live.
The gateway
One front door, locked to current standards.
-
TLS 1.3 only
Built, in alpha testingEvery connection to the platform uses TLS 1.3. A client that cannot speak it is refused; there is no older-protocol exception.
-
Approved algorithms only
Built, in alpha testingOnly encryption the Australian Signals Directorate approves. Anything else is refused, and automated tests check it.
-
Post-quantum ready
Built, in alpha testingThe gateway already offers encryption designed to resist future quantum computers, and our records are built so keys can be changed without moving data.
-
Mutual TLS between services
Built, in alpha testingServices prove who they are to each other with credentials that expire quickly. No service trusts a request because of where it came from.
-
Revocation
Built, in alpha testingA withdrawn credential stops working at the next check.
-
Limits at the edge
Built, in alpha testingLimits on request size and rate, and access logs that never record what you searched for or typed.
Identity
Every call says who it is for.
-
Short-lived tokens
Built, in alpha testingCalls between our services carry signed credentials that last minutes, checked at the front door and again by the service.
-
Delegated identity
Built, in alpha testingWhen PracticeIQ calls LexIQ for a person, that person's identity and permissions travel with the call as a signed claim.
-
Agents bound to the person
In developmentBy design, an AI agent's access is the overlap of the person's rights and the task it was given, never more; enforcement is in development.
-
Multi-factor sign-in
Built, in alpha testingSecond-step sign-in (security key or authenticator app) is built in the identity service and runs on our test server. It is required for firm administrators and operators, and by default for every member.
-
Your own identity provider
PlannedSign-in through a firm's own identity provider, for example Entra ID, before the first hosted firm.
-
No standing administrator access
Built, in alpha testingThe system administrator has no standing access to a firm's content; support access needs a time-limited grant from the firm and is recorded.
Isolation and encryption
Your firm, your database, a key per firm.
-
A database per firm
Built, in alpha testingEach firm's data lives in its own database. A request that cannot be tied to the right firm is refused, and tests prove that one firm cannot read another's data.
-
Client text sealed per firm
In developmentThe application encrypts client text with a key per firm, so a copy of the database or a backup shows no client text and one firm's key cannot open another's. Being rolled out step by step: client names, questions and answers are done.
-
Encrypted backups to two recipients
Built, in alpha testingEvery LexIQ backup is encrypted to a recovery key held offline and to a drill key that proves the backup restores each week. The other services follow.
-
No unencrypted copies at rest
Built, in alpha testingA LexIQ backup is encrypted as it is made and the unencrypted file is removed; existing unencrypted dumps were encrypted.
-
Erasure that is final
In developmentDestroying a firm's keys makes its sealed text unreadable once backups age out.
Confidentiality levels and local models
You decide where every question goes.
The firm administrator chooses a level, and a matter can be made stricter than the firm, never looser. The code enforces it: an outside call the level does not allow is refused.
| Level | What leaves your server | Status |
|---|---|---|
| Maximum | Nothing. Search runs locally; written answers come from a local model once it passes our evaluation, and until then the level returns cited passages. | Built, in alpha testing level enforced · In development local written answers |
| Enhanced | Search runs locally. A question and the passages needed to answer it go to a hosted AI provider outside Australia, under its retention terms. | Built, in alpha testing |
| Standard | Document text is processed by a hosted search provider when indexed, and questions and passages go to a hosted AI provider outside Australia. | Built, in alpha testing |
-
A warning before data leaves
Built, in alpha testingBefore a firm chooses Enhanced or Standard, it sees a data-sovereignty warning naming the providers, their regions and retention. Accepting it is recorded, and it is shown again whenever it changes.
-
Confidential matters
Built, in alpha testingMark a matter confidential and choose whether an outside call warns, is allowed or is blocked. A warning must be confirmed, and the confirmation is recorded.
-
Local models on your hardware
In developmentA local model service we build and maintain ourselves, running on the firm's own server or, in the cloud profile, on servers in Australia. It is running on our test server; the model for written answers is still being evaluated.
-
Provider terms checked by code
Built, in alpha testingA provider whose data terms we have not recorded is refused before any call is made, and every hosted call is logged before it leaves.
AegisIQ
A security monitor that tells you, and never strikes back.
AegisIQ watches every layer of the platform. Firms are informed about threats to them; the system administrator stays in control of every action.
-
Detection, measured first
In developmentDetections for credential attacks, scanning, floods, token abuse, tenant-isolation probing, data-exfiltration signals and misuse by AI agents. Each is measured against replayed attacks before it ships. The rules are built and start running live on our test server once the rule set is signed off.
-
Notifications for each firm
In developmentA notification inside PracticeIQ when a threat concerns your firm: what happened, the risk and impact to you, and what, if anything, to do. Built and on our test server; not yet live for firms, because the detection rules are not running there yet.
-
A monthly report for each firm
In developmentA monthly threat report covering only the events and actions that bear on your firm's risk, with no other firm's data. Built and on our test server; not yet live for firms.
-
Files scanned before anyone opens them
In developmentAegisIQ's scan service checks each uploaded file for malware and nothing is readable until the scan is clean. It runs on our test server; connecting it to every upload path is in progress.
-
Daily report and immediate alerts for the operator
In developmentThe system administrator receives a daily site-wide report and an immediate alert for any active threat. Alert grouping is built; the live channel is in progress.
-
Defensive only
In developmentAegisIQ never scans, probes, exploits or retaliates against systems we do not own. Containment, when built, is a closed list of reversible actions on our own systems: block at our boundary, keep evidence, report through proper channels.
-
A person approves
PlannedNothing AegisIQ recommends is actioned without the system administrator's approval, except short, reversible steps the administrator has enabled in advance. Patches are proposed for review and never merged by AegisIQ.
The engineering standard
Controls tracked against the Australian Government's Information Security Manual, with the gaps in writing.
We track our controls against the Australian Government's Information Security Manual and the Australian Privacy Principles; we have not been independently assessed. Our registers are self-assessments, not audits or certifications. This is where they stood on 25 Oct 2026; they change as work lands.
ISM guidelines for software development
Of the requirements that apply, about nine in ten are met or partly met; one is a gap, with the work that closes it named; one is still under consideration.
Australian requirements: security, privacy, consumer and profession rules
Of the requirements that apply, about two thirds are met or partly met; about a quarter are gaps, with the work that closes them named; about one in ten are still under consideration.
-
Safety properties are tests
Built, in alpha testingIsolation between firms, the confidentiality levels, read-only access and spending limits are automated tests, each proven able to fail. A failing one is fixed, never re-run until it happens to pass.
-
Every call metered
In developmentEvery request is metered today; the remaining kinds of work are being added so all usage is measured per firm and per matter.
-
No client text in logs
In developmentSecurity events carry metadata only, by design and by test. Platform-wide observability with a content deny list is planned.
-
Supply chain
Built, in alpha testingEvery third-party component pinned and checked, an inventory of the software and the encryption we use, and an automatic check that no secret is ever committed.
Privacy
Built around the Australian Privacy Principles.
Your firm is responsible for its clients' information; we build the controls you need to meet that responsibility. This is not legal advice about your obligations.
-
Data residency
PlannedThe cloud profile is designed to keep firm data in an Australian region, and the local modes keep it on your server. At the hosted levels, questions and passages go to providers outside Australia, as the warning says. No production hosting is running yet.
-
Notifiable data breaches
In developmentA breach runbook built on the OAIC's assess, contain and notify steps, and an audit that shows who accessed what and what left. The notice period from us to your firm is still to be set.
-
Export and erasure
Built, in alpha testingExport what you hold about a person and erase it on request, with the action recorded. Erasure happens only after the firm administrator types a confirmation.
-
Retention and legal hold
In developmentRetention periods with a seven-year floor for firm records, a review queue the firm administrator decides, and legal hold. Nothing is de-identified or deleted by retention until the firm administrator authorises it; copies such as cached answers expire on their own.
-
Consent before learning
PlannedThe system learns from a firm's work only after the firm's owner or administrator consents, and what it learns stays inside that firm.
-
Sharing only with consent
PlannedNothing learned from one firm reaches another firm without that firm's explicit consent.
What is not done yet
The open items, in plain view.
We do not put real client data on the platform while the security-critical items are open.
| Open item | Planned milestone |
|---|---|
| Sign-in through a firm's own identity provider (for example Entra ID) | Before the first hosted firm |
| Client text sealed with the firm's key in every remaining table | LexIQ Alpha 1, in progress |
| Fully local written answers at the Maximum level | After the local model passes our evaluation; in progress |
| AegisIQ's detection rules running live | Built; they start once the rule set is signed off |
| AegisIQ alerts running live | AegisIQ M1, target 26 November 2026 |
| Reversible containment with approvals and a kill switch | AegisIQ M2, target 10 December 2026 |
| Threat notifications and monthly reports live for each firm | Built on our test server; live once the detection rules run |
| A public test address and a dynamic security scan baseline | Platform foundations milestone; domain registered; needs DNS and certificate set-up |
| Backups that cannot be altered for their retention, on separate infrastructure | Open, no date set |
| Malware scanning of uploaded and imported files | Built (AegisIQ scan service, test server); wiring into every upload path in progress |
| Session timeouts and password rules to the current ISM | Built; the idle limit for members is longer than the ISM recommends |
| Independent review of the threat model and independent security testing | Planned, no date set |
| A system card for firms' due diligence | Planned, no date set |
| Customer terms, a vendor-to-firm breach notice period and legal review | Before the beta |
See also lessons from the industry, which grades what these controls would and would not have changed in real incidents.
Sources
Where the outside facts come from
Each source was read on 20 Oct 2026. Where we could not open a page directly, the entry says how it was read.
- Data breach notifications increase to all-time high in 2025, new NDB stats show
- Generative artificial intelligence: Practice Note SC Gen 23 and related material
- Use of Generative Artificial Intelligence Practice Note (GPN-AI)
- Statement on the use of artificial intelligence in Australian legal practice
Put your hardest confidentiality question to us.
Register your firm for a trial in the beta programme, or get occasional product updates as each open item closes.