engagement letters in 2025 — has anyone updated theirs for AI?
following up on a few threads about liability and AI. has anyone actually updated their engagement letters to address AI usage?
specific questions:
- do you disclose that you use AI tools in your workflow?
- do you disclaim liability for AI-generated errors?
- do you address clients using AI on their own (like anna's thread about the ChatGPT-prepared return)?
i've been putting it off but after seeing the thread about the client who filed with AI numbers and got hit with penalties, i think this needs to happen now.
happy to share our current template (in Dutch and English) if anyone wants a starting point.
3 replies
updated ours in January. here's the key additions:
1. Our use of AI (disclosure): "Firm may use artificial intelligence tools as part of its preparation and review processes. All AI-assisted work is reviewed and verified by qualified practitioners before delivery to Client."
2. Client's use of AI (disclaimer): "If Client has used AI tools to prepare any portion of the work product submitted to Firm for review, Client must disclose this at the time of submission. Firm's liability is limited to the specific items reviewed and does not extend to AI-generated work not presented for Firm's review."
3. AI limitation clause: "Firm does not warrant that AI tools are error-free. Firm's professional responsibility extends to the final work product delivered to Client, which reflects Firm's professional judgment regardless of tools used in preparation."
took about an hour to draft and get reviewed by our lawyer. well worth it.
rachel's template is excellent. i'd add one more clause that we included:
4. Data processing: "In the course of providing services, Firm may process Client data using third-party AI tools. Firm takes reasonable steps to protect Client data, including anonymising personally identifiable information where practicable. Client consents to such processing as part of this engagement."
this is particularly important for GDPR compliance in the EU/UK. if you're sending client data to OpenAI or Anthropic, even in anonymised form, you should have consent.
jan, happy to share ours too. one thing we added that rachel didn't mention:
escalation clause: "If during the course of engagement, Firm identifies that Client has filed tax returns or made regulatory filings based on AI-generated calculations without professional review, Firm will advise Client of any identified errors and recommend corrective action. Additional fees may apply for remediation work."
this protects you from the "i filed it myself with ChatGPT and now i need you to fix it" scenario. the remediation is a separate engagement with separate fees.
Sign in as a verified accountant to reply.
Sign in