This is a starting point, not a finished policy. Replace everything in [square brackets], delete what doesn't fit your practice, and have it reviewed by whoever advises your firm on IT, data protection and professional ethics. It is written for a small or mid-size Indian CA practice, and it is tool-neutral: it works for Claude, ChatGPT, Copilot, Gemini or any similar service. --- [Firm name]: Policy on the use of AI tools Version: [1.0] · Approved by: [partner name] · Effective from: [date] · Next review: [date, at most 12 months later] 1. Purpose and scope 1.1 This policy sets out how partners, staff, articled assistants and anyone working for [Firm name] may use artificial intelligence (AI) tools in professional work. 1.2 "AI tools" means any service that generates or analyses text, numbers, code or images on request, including chat assistants, coding assistants, AI features inside office, accounting or audit software, and browser extensions. 1.3 It applies to all client work, internal work and any device used for firm work. 2. Principles 2.1 Confidentiality comes first. Client information is confidential. Under the Chartered Accountants Act, 1949, disclosing information acquired in a professional engagement without the client's consent, other than as the law requires, is professional misconduct. Treat sending client data to an AI service as a possible disclosure. 2.2 Responsibility does not move. The member who signs or issues a work product is responsible for all of it, including anything an AI tool produced. "The AI said so" is never a reason. 2.3 Send the minimum. Give an AI tool only the data the task needs. Prefer methods where the tool writes a formula, macro or script that runs on our own computer, so the client data itself is not sent. 2.4 Verify before use. AI tools make confident mistakes, including invented sections, case law, rates and due dates. Every output is checked before it is relied on. 3. Approved tools and accounts 3.1 Only the tools listed in Annexure A may be used for firm work. To add a tool, ask [IT owner or partner]. 3.2 Use only firm-approved accounts for client work. Personal or free accounts must not be used with client information. 3.3 Before a tool is approved, [IT owner] checks and records: whether our data is used to train the provider's models, and how to turn that off; how long data is kept; where it is processed; and the provider's terms for business use. These are re-checked at each policy review, because plans and settings change. 4. What may be shared | Category | Examples | Rule | |---|---|---| | Green: no client data | Questions about law or standards, drafting templates, Excel formulas, writing a macro or script on dummy data | Allowed on approved tools | | Amber: client data, minimised | Column headers, a few masked sample rows, totals and counts; anonymised extracts | Allowed on approved tools, with client consent (section 5) and the steps in section 6 | | Red: never | Passwords, OTPs and portal or bank logins; Aadhaar and bulk PAN or KYC data; bank account details; personal data of individuals not needed for the task; material received under a confidentiality condition from a regulator, investigation or search; any other client's data | Not allowed on any AI tool | 4.1 When in doubt, treat data as Red and ask [partner]. 4.2 If personal data of individuals is involved, the Digital Personal Data Protection Act, 2023 may apply, and the provider's servers may be outside India. Check with [partner or DPO] before using such data. 5. Client consent 5.1 Client data (Amber) may be used only where the engagement letter, or a written confirmation from the client, allows the use of AI tools. A sample clause is in Annexure B. 5.2 For audits, investigations and other sensitive engagements, the engagement partner decides case by case, and records the decision in the file. 6. How to use AI tools on client work 6.1 Work from a separate folder for each engagement, holding copies (never originals) of only the files that job needs. 6.2 For data work, ask the tool for a formula, macro or script that runs locally, and check it on a small sample before running it on the full file. 6.3 Where the tool asks for permission to read files, run commands or connect to the internet, read each request before approving it. Refuse anything outside the engagement folder or that you did not expect. 6.4 Do not paste client data into a chat when the task can be done without it. 6.5 Tie the output back to the source: row counts, control totals and a sample of figures recomputed by hand. 7. Review and sign-off 7.1 Every section, rule, rate, due date, standard and case reference in AI output is checked against the primary source before use. 7.2 Anything going to a client, a regulator or for signature is reviewed by a second person, who is told that an AI tool was used. 7.3 AI output is not audit evidence by itself. Where an AI-assisted analysis supports an audit conclusion, the working papers show how its completeness and accuracy were tested. 8. Documentation 8.1 Where an AI tool was used on an engagement, the working papers record: the tool, the inputs (file names), what data was sent, the prompt or instructions, any script used, and the output. For example: "Prepared with [tool]. Inputs: [file names]. Script: [file name], saved in the job folder. Data sent: headers, 3 sample rows, counts. Reviewed by [name]." 8.2 Scripts and outputs needed as evidence are kept with the working papers under our usual retention policy. Scratch files are deleted when the job closes. 9. Not allowed - Using AI tools to give an opinion, sign a report or take a professional judgement on the firm's behalf. - Citing a section, notification, standard or case from AI output without reading the source. - Entering Red data (section 4) into any AI tool. - Using unapproved tools or personal accounts for firm work. - Presenting AI-written work to a client as reviewed when it has not been. - Using AI tools to create content that misleads, or that breaches ICAI's rules on advertising and solicitation. 10. Incidents 10.1 If client data is sent to an AI tool in breach of this policy, or a tool behaves unexpectedly with client data, tell [partner] the same day. Do not try to fix it quietly. 10.2 [Partner] decides on next steps, including telling the client, and whether the firm has notification duties under the Digital Personal Data Protection Act, 2023 and its rules. Take advice on this. 11. Training and review 11.1 Everyone covered by this policy reads it and signs Annexure C before using AI tools for firm work. 11.2 [Partner] reviews this policy at least once a year, and whenever ICAI issues guidance on AI or technology, or an approved tool changes its terms. --- Annexure A: Approved tools | Tool | Plan or account | Approved for | Training setting checked | Owner | Date | |---|---|---|---|---|---| | [Tool name] | [Firm business plan] | [Green and Amber] | [Off: date checked] | [Name] | [Date] | Annexure B: Sample engagement letter clause > We may use artificial intelligence tools approved by our firm to assist with parts of this engagement, > such as data analysis and drafting. We will limit the information shared with such tools to what the task > requires, use only services whose terms meet our confidentiality obligations, and review all output > ourselves. Responsibility for our work remains with us. If you do not want us to use such tools on your > information, or want to restrict their use, please tell us in writing. Annexure C: Acknowledgement I have read the [Firm name] AI Use Policy, version [1.0], and I will follow it. Name: ____________ Signature: ____________ Date: ____________ --- This template is general information, not legal or professional advice. Check it against current law, ICAI pronouncements and your firm's own obligations before adopting it.