The Singapore Law Gazette

PART 2: Using Generative AI in Legal Work

Supervision, Approval, and Responsibility in a Multi-Platform Environment

This article forms Part 2 of the Practice Support series. In Part 1, we examined how law firms can add value when clients generate their own AI-assisted drafts. This instalment turns to the operational core of the issue: how law firms in Singapore should supervise, approve, and govern legal work in an environment where clients, and increasingly firms themselves, deploy multiple generative AI tools.

The focus here is practical. Many firms are now working alongside external technology providers to build internal or client-facing legal AI platforms that integrate general purpose tools such as OpenAI’s ChatGPT, Anthropic’s Claude, Microsoft’s Copilot, and Google’s Gemini with specialist legal AI platforms such as Harvey, CoCounsel or Luminance.

The question is no longer whether these tools are used. The real questions are:

Who supervises these integrated hybrid platforms? Who approves their outputs? Who bears responsibility?

1. AI in Legal Workflows: Assistive Infrastructure, Not Professional Substitution

Clients are increasingly operating hybrid workflows:

• Business team drafts with ChatGPT.
• Technical team cross-checks with Gemini.
• Procurement team refines language with Copilot.
• Specialist clauses are generated using a bespoke LLM platform built by an external software provider.

Law firms are increasingly being asked to participate in this ecosystem, not merely to review documents, but to help architect the client’s legal AI platform.

This creates a new professional dynamic:

• AI generates.
• External technology providers integrate.
• Clients operationalise.
• Law firms supervise and assume accountability.

AI is therefore baseline infrastructure. But infrastructure does not displace professional judgment. It amplifies the need for it.

2. Supervisory Responsibilities of Lawyers in a Multi-AI Environment

Where multiple AI assistants are deployed, risk multiplies in subtle ways.

Different models may:

• Rely on different training data
• Apply different reasoning patterns
• Produce inconsistent clause interpretations
• Embed outdated case law A lawyer’s supervisory role becomes more complex, not less.

The Duty of Independent Review

Regardless of whether a draft originates from:

• A client’s internal AI platform,
• A third-party legal AI solution provider, or
• The firm’s own AI environment,

The supervising lawyer must:

1. Independently verify legal propositions.
2. Confirm regulatory and legal compliance under Singapore law.
3. Ensure internal consistency.
4. Validate contextual suitability to the transaction.

AI output is never self-validating.

Specific Supervisory Risks

In practice, lawyers must be alert to:

• Hallucinated case citations
• Non-existent statutory provisions
• Outdated regulatory thresholds
• Clause inconsistency across AI platforms

When multiple AI assistants are used in parallel, one may “correct” another but neither may be correct.

Supervision must therefore be structured, not casual.

3. Working with External Legal AI Service Providers

An emerging model in Singapore involves:

• A law firm collaborating with a software solutions provider specialising in LLM integration.
• The provider builds a customised legal AI platform for the client.
• The platform integrates multiple AI assistants.
• The law firm defines legal guardrails and governance standards.

In this tripartite structure (Client – Tech Provider – Law Firm), clarity of roles is critical.

Allocation of Functions

Technology Provider

• System architecture
• Model integration
• Prompt engineering optimisation
• Security infrastructure

Law Firm

• Legal knowledge design
• Risk allocation templates
• Compliance parameters
• Escalation thresholds
• Governance oversight

The technology provider builds the engine.

The law firm focuses on substantive laws and provide contextualised legal analysis

Contractual Safeguards

When working alongside external providers, law firms should ensure:

• Clear delineation of liability
• Data ownership clarity
• Confidentiality undertakings
• Security standards alignment
• Incident notification obligations

AI governance is not only about reviewing the final legal outputs. It is about oversight within a structured protocol.

4. Approval and Sign-Off Frameworks Within Law Firms

In a multi-AI workflow, ad hoc or “light touch” review is insufficient.

Firms should implement structured approval protocols.

Tiered Internal Approval Model

Level 1 – Associate Review

• Clause accuracy
• Internal consistency
• Statutory alignment
• Identification of anomalies

Level 2 – Senior Associate / Counsel Review

• Risk allocation benchmarking
• Market standard comparison
• Negotiation positioning
• Escalation trigger assessment

Level 3 – Partner Sign-Off

• Strategic exposure review
• Contextual legal and business risk assessment
• Regulatory sensitivity analysis
• Board-level suitability

No AI-generated document should bypass professional hierarchy.

Documentation of Approval

Every AI-assisted engagement should record:

• Which AI tools were used
• Scope of review performed
• Assumptions validated
• Key risk deviations identified
• Escalation decisions made

Auditability protects both firm and client.

5. Professional Responsibility Cannot Be Delegated

AI tools do not bear professional liability.

They are not officers of the court.
They are not regulated by the Law Society.
They do not carry professional indemnity insurance.

The signing partner remains responsible.

Regulatory and Ethical Dimensions

Lawyers must ensure:

• Competent supervision of technology used
• Protection of client confidentiality
• Avoidance of misleading reliance on AI outputs
• Transparent communication with clients regarding AI involvement

AI is just a tool to facilitate lawyering. Professional responsibility will always remain with the supervising lawyer and the law firm.

6. AI Governance: Personal Data and Confidentiality

Where clients deploy AI platforms integrating tools such as ChatGPT, Claude, DeepSeek, Copilot and Gemini, data flow becomes a central governance issue.

Questions include:

• Is data transmitted outside Singapore?
• Is the client’s data used for model retraining?
• Are prompts captured for continuing learning?
• Are cross-border transfers occurring?

Firms must ensure that personal data handling complies with Singapore’s data protection framework and client confidentiality obligations.

Governance Safeguards

A robust AI governance framework should include:

1. Data Segmentation

• Sensitive data redaction protocols
• Use of anonymisation layers
• Separation of public and private datasets

2. Access Controls

• Role-based permissions
• Multi-factor authentication
• Logging of prompt usage

3. Audit Trails

• Record of AI interactions
• Version history tracking
• Documented overrides

4. Escalation Thresholds

• Mandatory legal review for high-value contracts
• Automatic escalation for regulatory-sensitive matters
• Human sign-off for cross-border transactions

Governance must be embedded into the platform, not applied retrospectively.

7. Practical Safeguards for Law Firms Building Client AI Platforms

When assisting clients in implementing a multi-AI legal system, firms should:

A. Define Permitted Use Cases

Specify:

• Types of documents allowed for AI drafting
• Categories excluded (e.g., highly regulated filings)
• Value thresholds for mandatory legal review

B. Create Approved Prompt Libraries

Develop:

• Structured prompt templates
• Risk-calibrated clause instructions
• Fallback language banks
• Pre-approved negotiation ranges

C. Embed Risk Scoring

Introduce:

• Automated clause risk flags
• Red-amber-green risk matrices
• Built-in compliance reminders

D. Establish Clear Escalation Rules

For example:

• Indemnity caps exceeding X threshold
• Limitation clauses removing fraud carve-outs
• Non-compete clauses exceeding statutory limits
• Data processing clauses involving cross-border transfers

AI should trigger supervision, not bypass it.

8. Practical Checklist for Practitioners

When Reviewing AI-Generated Drafts

• Was the draft generated using multiple AI tools?
• Are there inconsistencies between sections?
• Are statutory references current?
• Are case citations verified independently?
• Does the allocation of risk align with market norms?

Supervision Checklist

• Independent validation performed?
• Regulatory framework checked?
• Client-specific context verified?
• Industry-specific compliance assessed?
• Cross-border implications considered?

Approval and Sign-Off Checklist

• Senior review completed?
• Escalation thresholds triggered?
• Risk memo prepared (if required)?
• Documentation recorded?
• Client informed of material risk adjustments?

AI Governance Checklist (Client Platform Context)

• Data protection safeguards implemented?
• Confidential information anonymised?
• Clear data retention policy?
• Audit logs enabled?

9. The Emerging Hybrid Model: Lawyers as Supervisors of Intelligent Systems

Legal practice in Singapore is moving towards a hybrid structure:

• AI generates first cuts.
• Technology providers integrate systems.
• Clients operationalise platforms.
• Law firms supervise, approve, and remain accountable.

The lawyer’s role is evolving from drafter to:

• Systems supervisor
• Governance architect
• Risk approver
• Accountability bearer

AI increases drafting efficiency. It simultaneously increases the need for structured oversight.

Conclusion: Accountability Is the Anchor

The integration of multiple AI tools into legal workflows is now a practical reality. Multi-assistant platforms will only grow more sophisticated.

But three principles remain constant:

1. Supervision is mandatory.
2. Approval must be structured.
3. Responsibility cannot be delegated.

AI’s legal cognitive and reasoning capabilities are constantly improving at rapid pace.

Lawyers must add strategic value and move up the value chain.

And lawyers overseeing smarter AI systems will remain accountable to clients.