Doc 09 — AI / Automation Terms

1. What these AI Terms cover

1.1 These AI / Automation Terms ("AI Terms") govern your use of the artificial-intelligence and automation features of the Platform ("AI Features"). They form part of, and are incorporated into, the Platform Terms of Service (see Doc 02 — Platform Terms of Service). Capitalised terms not defined here have the meanings given in Doc 02.

1.2 If there is any conflict between these AI Terms and Doc 02 on the subject of AI Features, these AI Terms prevail for that subject matter. The Engagement / Terms of Business (Doc 03 — Singapore Terms of Business) prevails over both for Professional Services.

1.3 Application to Participating Firms. Where these AI Terms are incorporated into the Participating Firm Terms (see Doc 11 — Participating Firm Terms), references to "you" are read as references to the Participating Firm. The Participating Firm must not treat any AI Output as discharging, or as a substitute for the performance of, any professional, statutory or regulatory obligation it owes, and remains solely responsible for every professional decision (mirroring Doc 11 clause 9).

2. What the AI Features do

2.1 The Platform uses AI Features to assist the people who serve you. AI Features may, depending on your services and configuration:

(a) classify inbound requests and communications (for example, routing an email enquiry to the right workflow);

(b) extract information from documents and communications you provide, to pre-fill forms and records for human checking;

(c) generate summaries of cases, documents or correspondence;

(d) assist in preparing documents and provide drafting assistance;

(e) recommend workflow steps or next actions to staff and Professionals;

(f) assist in triaging risk-related information (see clause 6); and

(g) assist in responding to support queries.

2.2 AI Features are assistive, not autonomous. They produce suggestions, drafts, classifications and pre-filled records for review by you, by DPW's staff, or by the Participating Firm and its Professionals. No AI Feature independently forms an engagement, approves an identity check, gives advice, or files anything with any government authority.

2.3 AI-assisted routing among Participating Firms. The Platform may use rules, software and AI Features to assist in matching, routing, prioritising and allocating work among Participating Firms, based on the objective platform criteria stated in Doc 12 — Marketplace / Professional Network Terms. Any such automated or AI-assisted routing or allocation is an operational platform function: it selects which Participating Firm receives an opportunity to act. It is not a professional, legal or statutory decision, does not transfer professional responsibility to DPW or to any software, and never replaces the human professional decisions described in Doc 20 Section 2.

3. What AI output is NOT

3.1 Output produced by an AI Feature ("AI Output"):

(a) is not legal, accounting, tax, corporate secretarial or any other professional advice, and does not create any adviser–client or professional relationship;

(b) is not a professional approval, compliance determination, identity-verification decision or clearance of any kind;

(c) is not authority to file, and does not cause, permit or constitute any filing with ACRA, IRAS or any other authority; and

(d) does not approve any engagement or trigger any filing merely because software — including any Third-Party Service Provider — returns a "clear", "pass", "verified", "low risk" or similar result.

3.2 No automated step approves anything: every approval and every filing is a human professional decision of the Participating Firm, and government acceptance rests solely with the relevant authority (see Doc 20 Section 2 for the full decision chain).

4. Accuracy and safeguards

4.1 AI Output may be incomplete, inaccurate, out of date or wrong, including in ways that appear plausible. Language models can misread documents, miss information, or generate content that has no basis in your materials.

4.2 You must review AI Output before relying on it, and the Participating Firm's Professionals must review AI Output before any professional decision or filing based on it. AI Output is a starting point for human work, not a substitute for it.

4.3 The Platform applies deterministic (rule-based, non-AI) safeguards and human-review gates around AI Features, including mandatory human approval of identity checks by authorised staff roles and manual professional filing (see Doc 20). Automated steps operate behind kill-switch controls that allow DPW to suspend an automated step and revert it to manual handling.

4.4 These safeguards reduce, but do not eliminate, the risk of AI error. They do not convert AI Output into advice or approval.

5. Data handling in AI Features

5.1 Prompts sent to AI models pass through a mandatory sanitisation gateway operated by the Platform before leaving it.

5.2 For most AI tasks, identity data is redacted before processing: NRIC/FIN numbers, passport numbers, email addresses, phone numbers, dates of birth and addresses are replaced with placeholders. Names are not redacted for these tasks.

5.3 One category of task — extraction of information from documents and communications — processes unredacted identity data, because redaction would defeat its purpose. This task is subject to a technical deployment guard that restricts it to a single designated AI model provider and blocks (fails closed) any attempt to route raw identity data to any other provider.

5.4 Data processed by AI Features may be processed by the AI model providers listed in the Subprocessor and Service Provider Notice (see Doc 06 — Subprocessor and Service Provider Notice), which also records each provider's location and transfer mechanism. How personal data is handled in AI Features is further described in the AI section of the Privacy Notice (see Doc 04 — Privacy Notice).

5.5 Training. DPW does not itself use your content or personal data to train AI models. Restrictions on any AI model provider's use of submitted data for its own model training, and on provider-side retention, are [EVIDENCE_GATED — subject to provider contractual confirmation]; DPW will not enable production AI processing of a data category with a provider unless the applicable contractual data-protection terms with that provider are in place. DPW does not represent that data submitted to AI model providers is never used for training beyond what those contractual terms provide.

6. Automated risk classification and your right to human review

6.1 Where screening of a person's name against sanctions, politically-exposed-person and adverse-media databases returns no match, the Platform automatically classifies that screening result and assigns an indicative risk rating, to assist the Participating Firm. Any match, uncertainty or error creates a mandatory task for human review; the automated step never resolves it, and screening errors fail closed to human handling.

6.2 An automated classification is an input to, never a substitute for, human decisions. Every approval of an identity check, every acceptance of an engagement and every filing requires a decision by an authorised person of the Participating Firm (with dual approval for certain roles). This design is intended to align with the PDPC's Advisory Guidelines on the use of personal data in AI recommendation and decision systems.

6.3 Right to human review. You may ask, via the contacts in clause 12, that any material AI-assisted output affecting you — such as a risk classification, an extraction used to populate your records, or an AI-assisted decision input — be reviewed by a human. See also Doc 15 — KYC / AML / Identity Verification Notice for how this operates in identity verification.

7. Your responsibilities when using AI Features

7.1 You must:

(a) review AI Output for accuracy and completeness before using or submitting it;

(b) not input personal data or confidential information of another person unless you have the authority or consent to do so;

(c) not rely on AI Output as legal, tax, accounting or other professional advice, or as any form of approval or clearance; and

(d) not misuse AI Features, including by attempting prompt injection, attempting to extract system prompts or other users' data, generating unlawful content, or otherwise circumventing safeguards, as further set out in the Acceptable Use Policy (see Doc 08 — Acceptable Use Policy).

7.2 We may suspend or restrict your access to AI Features for misuse under Doc 08 and Doc 02.

8. Beta and experimental features

8.1 We may make AI Features available on a beta, preview, pilot or experimental basis, identified as such. Beta features are provided "as is" for evaluation, may be less accurate or reliable, may change or be withdrawn at any time, and should not be used for any decision of consequence. The safeguards in clauses 3, 4 and 6 continue to apply to beta features: no beta feature approves anything or files anything.

9. Changes to AI Features

9.1 We may change AI Features, including the underlying models and providers (which will be reflected in Doc 06), to improve quality, safety, cost or compliance.

9.2 We will notify you of material changes to AI Features — including any material change to what AI Features do with your data — in the manner set out in Doc 02.

9.3 We will not remove or materially reduce the human-review gates described in clause 4.3 and Doc 20 (human approval of identity checks; manual professional filing) without prior notice to you, and never in a way that conflicts with the Participating Firm's legal obligations.

10. Liability

10.1 AI Features are part of the Platform Services, and DPW's liability in connection with them is subject to the liability provisions of Doc 02, including the cap and the carve-outs (for death or personal injury caused by negligence, fraud, and other liability that cannot be excluded or limited under Singapore law) set out there.

10.2 To the maximum extent permitted by law (including the Unfair Contract Terms Act 1977), DPW is not liable for loss arising from your reliance on AI Output that you have not reviewed, or that you have used contrary to clauses 3, 4 or 7 — for example, treating AI Output as professional advice or as an approval. This clause does not exclude liability for DPW's own breach of these AI Terms, and does not affect the carve-outs referred to in clause 10.1 or any non-excludable rights under the Consumer Protection (Fair Trading) Act 2003.

10.3 The Participating Firm's responsibility for Professional Services, including its professional review of AI-assisted work, is governed by Doc 03, not by these AI Terms.

11. Related documents

11.1 See: Doc 02 — Platform Terms of Service (contract framework, liability); Doc 04 — Privacy Notice (AI section; personal data); Doc 06 — Subprocessor and Service Provider Notice (AI model providers); Doc 08 — Acceptable Use Policy (prohibited conduct); Doc 15 — KYC / AML / Identity Verification Notice (identity verification); Doc 20 — Professional Review and Filing Authorisation (human decision chain).

12. Contact

12.1 Questions about AI Features or these AI Terms, or requests for human review under clause 6.3: privacy@csfile.ai; data protection matters: dpo@csfile.ai (see Doc 04 — Privacy Notice).