A simple security checklist before using AI on client matters

Before a lawyer pastes a clause, uploads a bundle or connects an AI tool to the document system, the firm should be able to answer ten practical questions. This checklist gives Malaysian firms a workable starting point.

AI security does not begin with a long policy. It begins at the point where a lawyer decides what information to give a tool, who can see it and what happens to it afterwards.

For a 30–50 person Kuala Lumpur firm, the safest approach is not to ban AI or approve every product. It is to define a small approved environment, keep client data out unless the use has been cleared, and make one lawyer accountable for the final work.

Malaysia's Personal Data Protection Act 2010 (PDPA), as amended in 2024, applies to personal data processed in commercial transactions. Its security and retention principles matter when an AI service receives names, contact details, NRIC numbers, financial information, evidence or employee data. Since 2025, data processors also have direct obligations under the Security Principle, and qualifying organisations must appoint a data protection officer.

The Malaysian Bar's July 2025 GenAI advisory is deliberately cautious. It says lawyers should not disclose information subject to solicitor-client confidentiality or legal professional privilege, should avoid putting personal or client-specific data into GenAI tools, and must subject outputs to human oversight. That is the correct baseline.

The pre-use checklist

1. Is the material confidential or privileged?

Assume that client instructions, correspondence, draft advice, evidence and matter strategy are confidential. Do not paste or upload them to a general consumer AI tool.

Privilege is context-specific, and disclosure to a provider can create avoidable arguments about whether protection has been preserved. Client consent should not be treated as a substitute for proper controls. If the task can be done with a fictional or fully anonymised example, use that instead. Removing a name is not enough if the remaining facts identify the person or matter.

2. Will the provider use the data for training?

Check the contract, privacy terms and product settings, not the salesperson's summary. Ask whether prompts, files, outputs and feedback are used to train or improve any model, reviewed by humans, or retained for abuse monitoring.

The acceptable answer for client work is usually a binding no-training commitment for your firm's data. Confirm whether the commitment applies to every feature, including plug-ins, browsing, voice, support tickets and optional feedback.

3. Who can access the tool?

Use firm-managed accounts with single sign-on and multi-factor authentication. Disable personal accounts for client work. Give access by role, remove leavers promptly, and review access at least quarterly.

Do not share logins. Restrict administrator rights. If the product allows public links, external sharing or third-party connectors, switch them off by default.

4. Are matters isolated?

A user working on Matter A should not be able to retrieve content from Matter B. Test this rather than relying on a statement that the system is "enterprise-grade".

Use separate workspaces or repositories, matter-based permissions and ethical walls where required. Check how the tool indexes documents and whether retrieval can cross teams, clients or offices. Run a simple red-team test using dummy matters before approval.

5. What is retained, and how is it deleted?

Map every copy: prompts, uploads, outputs, chat history, embeddings, indexes, logs and backups. Set a retention period that matches the purpose and the firm's records policy. The PDPA Retention Principle requires personal data not to be kept longer than necessary.

Confirm that deletion can be initiated by the firm, covers derived data, and reaches backups within a stated period. Obtain an export and deletion process for contract termination. "Deleted from the user interface" is not the same as deleted from the service.

6. Which subprocessors and countries are involved?

Obtain the provider's current subprocessor list and the locations used for hosting, support, model inference and backups. Require notice before material changes.

A transfer to servers or support personnel outside Malaysia must satisfy section 129 of the PDPA. The 2025 cross-border transfer guideline places practical emphasis on identifying the applicable transfer condition, assessing protection, applying contractual and technical safeguards, and keeping records. Do not assume that a well-known cloud brand resolves the transfer analysis.

7. Are useful logs available?

The firm should be able to see who logged in, what was uploaded, which matter was involved, which model or feature was used, what was shared externally, and what an administrator changed.

Protect logs from ordinary users and retain them long enough to investigate an incident. Avoid collecting the full content of every prompt in a central log unless that is necessary; a security log can itself become a concentrated store of client material.

8. What happens when something goes wrong?

Put the AI service into the firm's incident response plan. Name the internal contact, provider contact, decision-maker and backup. The contract should require rapid notification, preservation of evidence, cooperation and clear responsibility for containment.

Under the amended PDPA, notifiable personal data breaches must be reported to the Commissioner as soon as practicable and within the prescribed process; affected individuals must also be notified where significant harm results or is likely. Build an internal escalation window measured in hours, not days. Test the plan with a lost account, an unintended upload and cross-matter retrieval.

Name a qualified lawyer for every client-facing output. That lawyer must verify quotations, facts, calculations, authorities and citations against primary or trusted sources. AI output is a draft, not evidence of accuracy.

Keep a proportionate record of the tool used, the review performed and the final approver. Do not let an AI system make the final call on advice, privilege, filing, settlement or a client's rights.

10. Is the pilot narrow enough?

Start with one approved tool, a small user group and low-risk tasks using public, synthetic or properly anonymised material. Suitable early uses include formatting, training materials, generic checklists and first-pass structure.

Set a start and end date, success measures, prohibited data, named owners and a stop mechanism. Review incidents, near misses and user behaviour before widening access. A pilot should test controls, not quietly become production.

A sensible first step

Choose one real workflow and complete this checklist with the managing partner, IT lead, data protection officer or privacy lead, and a lawyer from the relevant practice group. Record the decision and its limits.

Auxilator can help Malaysian firms turn that review into a short AI use policy, vendor questionnaire and controlled pilot. The objective is straightforward: use AI where it is useful, without weakening the duties clients already rely on.

Notes

  1. Malaysian Bar, Circular No 242/2025, "Updates on the Use of Generative Artificial Intelligence in Legal Practice": https://www.malaysianbar.org.my/cms/upload_files/document/Circular%20No%20242-2025.pdf
  2. Personal Data Protection Commissioner, Personal Data Protection Act 2010 [Act 709]: https://www.pdp.gov.my/ppdpv1/en/akta/pdp-act-2010-en/
  3. Personal Data Protection Commissioner, Personal Data Protection (Amendment) Act 2024: https://www.pdp.gov.my/ppdpv1/en/akta/personal-data-protection-amendment-act-2024/
  4. Federal Gazette P.U. (B) 522, commencement dates for the 2024 amendments: https://www.pdp.gov.my/ppdpv1/wp-content/uploads/2024/12/PENETAPAN-TARIKH-PERMULAAN-KUAT-KUASA-2.pdf
  5. Personal Data Protection Commissioner, Personal Data Protection Standard 2015: https://www.pdp.gov.my/ppdpv1/en/akta/personal-data-protection-standard-2015/
  6. Personal Data Protection Commissioner, Data Breach Notification Guideline (25 February 2025): https://www.pdp.gov.my/ppdpv1/wp-content/uploads/2025/08/GP_DBN_ENG.pdf
  7. Personal Data Protection Commissioner, Appointment of Data Protection Officer Guideline (25 February 2025): https://www.pdp.gov.my/ppdpv1/wp-content/uploads/2025/08/GP_DPO_ENG.pdf
  8. Personal Data Protection Commissioner, Cross-Border Personal Data Transfer Guideline No. 3/2025: https://www.pdp.gov.my/ppdpv1/wp-content/uploads/2025/08/GP_CBPDT_EN.pdf
  9. Department of Personal Data Protection, Data Protection by Design Guideline (2026): https://www.pdp.gov.my/ppdpv1/wp-content/uploads/2026/04/Data-Protection-By-Design-Guideline-DpbD.pdf
  10. MOSTI, National Guidelines on AI Governance and Ethics: https://www.mosti.gov.my/wp-content/uploads/2024/09/NATIONAL-GUIDELINES-OF-AIGE-20241118.pdf

Trusted and used by
a leading U.S. law firm.

End-to-end encrypted · Your data never trains other models · Matter-level client isolation

Auxilator is a product of Mustard Seed Credo Limited Company. © 2026 All rights reserved.