AI in Courts 2026: Why CAs Should Care About Disclosure and Evidence
The Supreme Court of India's draft Regulations for Use of Artificial Intelligence in Courts, 2026 are aimed at courts, not CA firms. But the principles matter for chartered accountants: AI must remain assistive, human responsibility stays with the professional, AI-assisted submissions may need disclosure, and nobody can use "the AI did it" as a defence for false or misleading material.
Facts checked: 26 August 2026 against the Supreme Court draft notice as reproduced by legal publishers, Indian Express reporting, and ICAI's AI background-material listing.
This is not a reason for panic. It is a useful preview of how courts, regulators and clients are likely to think about professional AI use.
Important status note: this is a draft judicial-AI framework, not a notified rule for CA firms. The reason it belongs in a CA firm's risk reading list is that it captures the direction of travel: AI assistance is acceptable only when human responsibility, disclosure and auditability remain intact.
What the draft court AI rules say
The draft framework is built around a few simple principles:
| Principle | Meaning |
|---|---|
| Human primacy | AI cannot decide judicial outcomes by itself |
| Assistive use | AI supports court processes but does not replace human judgment |
| Disclosure | Material AI assistance in submissions may need to be disclosed |
| Explainability | Opaque tools face scrutiny, especially in high-risk contexts |
| Audits and registers | Approved AI systems need records, audits and incident tracking |
| Responsibility | False or misleading AI-generated material remains the user's responsibility |
For CAs, the last two are the most important: records and responsibility.
Why this matters for CA work
Chartered accountants produce documents that often end up in legal or regulatory settings:
- audit reports,
- certificates,
- valuation reports,
- forensic findings,
- expert summaries,
- tax submissions,
- reply drafts,
- management letters,
- and working papers reviewed during disputes.
If AI helped prepare any of those, the professional risk question becomes: can you explain what was used, what it produced, how it was verified and who accepted responsibility?
The CA version of the court AI principles
1. AI output is not evidence by itself
An AI-generated summary of a ledger, contract or bank statement is not the source document. It is a derived output.
The audit file should retain:
- original source,
- data extract,
- transformation logic,
- AI prompt or instruction summary,
- AI output,
- auditor changes,
- and reviewer sign-off.
That is the minimum chain from evidence to conclusion.
2. Human responsibility cannot be outsourced
The draft court regulations make the same point audit standards already imply: AI can assist, but the human decision-maker remains responsible.
For CAs:
- the auditor signs the audit report,
- the valuer signs the valuation,
- the certifying CA signs the certificate,
- the partner signs the engagement conclusion.
The tool does not sign. The professional does.
3. Disclosure will become normal
The court draft discusses disclosure where AI materially assists documents, pleadings or evidence. CA work is not automatically under that rule, but the direction is clear.
In high-risk work, firms should be ready to disclose:
- whether AI was used,
- what type of tool was used,
- whether client data was processed,
- whether the tool was public, enterprise, local or vendor-hosted,
- what verification was performed,
- and who reviewed the output.
This does not mean every email needs an AI footnote. It means material professional outputs should have an internal AI-use record.
4. Explainability beats black-box scoring
If a tool says "high risk" but cannot explain why, the output is weak.
A defensible CA workflow should show:
- the rule breached,
- the data field used,
- the threshold applied,
- the exception generated,
- and the auditor's response.
This is why deterministic rules and audit trails matter more than a polished AI narrative.
Where CAs are most exposed
Certificates
Turnover, net worth, fund utilisation, source-of-funds, working capital and RERA certificates must be based on records. AI can draft wording or reconcile schedules, but the certificate should not rely on unsupported AI inferences.
Forensic and investigation work
Fraud indicators, related-party links, round-tripping patterns and bank-statement anomalies may be surfaced by AI. The final report needs source evidence, not just model output.
Expert opinions and court-facing reports
If a CA report may be filed in court or arbitration, AI assistance should be documented internally and disclosed where required by the forum or counsel.
Tax replies and appeal drafts
AI is useful for drafting facts and grounds, but fake citations or hallucinated case law are a direct professional-risk event. Every case reference must be verified.
A practical AI evidence log
For material professional outputs, keep a simple log:
| Field | What to record |
|---|---|
| Engagement | Client / matter / period |
| Output | Certificate, report, working paper, reply, memo |
| Tool used | CA GPT, Claude, local model, vendor platform, internal tool |
| Data used | Redacted facts, client ledger, public law, uploaded documents |
| Purpose | Drafting, classification, reconciliation, summarisation |
| Verification | Source checked, reperformance, reviewer review |
| Changes made | Material edits by CA |
| Final owner | Partner / manager responsible |
This is not bureaucracy. It is the future audit trail of AI-assisted professional work.
Frequently Asked Questions
Do the 2026 AI in Courts draft regulations apply directly to CA firms?
No. They are a draft framework for court processes. But they are relevant because they show how Indian institutions are thinking about AI: human responsibility, disclosure, explainability, data protection and auditability.
Should CAs disclose AI use in every document?
Not necessarily. Routine drafting assistance does not need to be treated the same as AI-assisted evidence or expert work. But material professional outputs should have an internal AI-use record, and court-facing submissions should follow the relevant forum's disclosure rules.
Can AI-generated output be used as audit evidence?
AI output by itself is not source evidence. It can help analyze or summarize evidence, but the file still needs the original record, method, exception list and auditor conclusion.
What is the biggest risk for CAs using AI in legal or regulatory work?
The biggest risk is unsupported confidence: fake citations, unverified summaries, black-box risk scores and polished language that hides weak evidence. Verification and audit trail are the controls.
Sources: ICAI AI background materials, Indian Express on Supreme Court draft AI regulations, Indian Express explainer on court AI rules, LiveLaw analysis, public reproduction of draft provisions.
Try CORAA -> Audit trails, source-linked exceptions and reviewer sign-off for AI-assisted audit work. See pricing · DPDP-safe prompt library · AI hallucination controls.