AICA Level 1, 2 and 3: ICAI's AI Certification, Reviewed (2026 Syllabus)

AICA stands for Artificial Intelligence for Chartered Accountants. It is the Institute of Chartered Accountants of India's own AI certification programme, launched in 2024 and structured across three levels.

AICA is a certification, and is distinct from CA GPT — ICAI's AI tools platform at ai.icai.org. People often search for one and mean the other. The certification teaches; the platform does the work.

For a role-by-role comparison across the main institutional options, see CA GPT vs AICA vs ISCA AI Fluency.

Two things have changed materially in 2026, and most of what is written about AICA has not caught up with either.

First, Level 2 is no longer an "AI awareness" course. The current Ver 3.0 syllabus teaches Python, full-stack app deployment, Android APK builds, local model execution with LM Studio, workflow automation with n8n, and agentic AI. It is closer to a developer bootcamp for CAs than to a CPE seminar. If your picture of AICA is slide decks about what machine learning is, that picture is two versions out of date.

Second, Level 3 was launched at ICAI's AI Innovation Summit on 26 June 2026 — and as of today, ICAI's own live Level 2 course document still lists it under "Future Roadmap" as something the Institute "plan to Introduce ... at later stage." Both things are true at once, and the gap between them is the single most useful thing to understand before you plan a certification path.

For practising CAs deciding whether to enrol — and for partners deciding whether to fund staff enrolment — the honest questions are: what does AICA actually cover, what does it skip, and is it worth the time and money compared to alternatives?

This post is the unvarnished review. Level 2 details below are taken from ICAI's own published course document, AI for Chartered Accountants-Level-2 Ver 3.0, at ai.icai.org — not from secondary reporting.

AICA Level 1 — what it is

• Format: 3-day in-person + hybrid program at ICAI branches across India
• Cost: ₹5,000 + GST per participant (~₹5,900 total)
• Frequency: Multiple batches per year. Recent batches B635 (Pune), B680 (Gurugram) March-April 2026
• Eligibility: Open to ICAI members and students
• Curriculum (approximate based on public announcements):
  • Introduction to AI fundamentals — what AI is, ML basics, generative AI overview
  • AI in audit — use cases at a high level
  • Demos of CA GPT (ICAI's in-house AI platform for CAs) and the use-case catalogue at ai.icai.org — now 150+ GPT-based tools
  • Hands-on exercises with prompt engineering
  • Discussion of DPDPA + AI confidentiality concerns
  • Q&A and certification

Honest assessment of Level 1:

Strengths:

• Affordable — ₹5,000 is accessible for individual practitioners
• ICAI brand — adds credibility to the CV; some clients value the credential
• Networking — meeting other CAs interested in AI; builds peer relationships
• Hands-on — actual exposure to CA-GPT, prompt engineering, real tools
• Curated material — material is vetted by ICAI committees; less risk of misleading content

Weaknesses:

• Surface-level depth — 3 days isn't enough to develop genuine proficiency
• CA-GPT-centric — significant time on ICAI's own platform, which is one option among many. Less coverage of Claude, ChatGPT, Perplexity, Grok comparison.
• Limited audit-execution focus — covers AI fundamentals + use cases, not the deep "how do I use AI in my next tax audit" workflow
• Theoretical — the 30-50 hours of practice / pilot work needed to actually internalise AI in your workflow happens AFTER the certification, on your own
• CPE-equivalent value uncertain — counts toward CPE hours but no career path advantage independent of demonstrated practical use

Who should do AICA Level 1:

• ✓ Junior CAs (articled / 1-3 years experience) wanting structured AI introduction
• ✓ Mid-career CAs without prior AI exposure seeking baseline credibility
• ✓ Partners wanting to signal AI literacy on their CV
• ✓ Staff at firms where the partner expects AI literacy as a baseline skill

Who should skip / supplement:

• ✗ CAs already actively using ChatGPT / Claude / CORAA in practice — Level 1 won't teach you much you don't know
• ✗ Partners deciding firm-level AI strategy — Level 1 is individual literacy, not firm strategy
• ✗ Practitioners with engineering / technical backgrounds — depth is below their level

For those skipping Level 1, consider:

• CORAA AI Lab — practical guide + 30+ prompts
• Adopting AI in Audit: 7-rule framework
• Claude for Indian Audit Work

AICA Level 2 — what it is

Stated course objective: "Enabling Chartered Accountants as AI Consultants by Implementing AI at Client Offices." That framing matters — Level 2 is not aimed at making you a better user of AI in your own audits. It is aimed at making you someone who builds and deploys AI tools for clients.

• Prerequisite: AICA Level 1 certificate. CAs only — unlike Level 1, students are not eligible
• Format: 5 days, 10 AM–5 PM — 2 days online on Zoom (max 200 participants), then 3 days physical (max 50)
• Cost: ₹10,000 + GST
• CPE hours: 30 structured hours
• Certificate: issued roughly 60 days after the batch, capstone submission and review

The Ver 3.0 syllabus

An online orientation runs 5–10 days before the batch starts.

The Tally connector module is the most India-specific part of the syllabus, because it points at direct accounting-system interaction rather than generic prompting. For the practical implications, see ICAI's Tally MCP Server and what it means for CA firms.

Assessment is out of 100. Daily objective tests on the Digital Learning Hub carry 50 marks — 10 marks each day, 5 PM to 10 PM, pass mark 5 out of 10, two attempts allowed. The individual capstone project carries 30 marks with a pass mark of 15, submitted within 10 days of the course as a single ZIP containing the project summary, prompt files, example files and executables — plus a video of you presenting it, showing both your face and the technical content, uploaded unlisted to YouTube or Drive.

Miss a daily test and ICAI reassigns you to a fresh online-only batch, where you retake all five days of tests. Miss Day 1 entirely and your registration is cancelled outright.

Practical requirements worth knowing before you register: you must bring your own laptop (Windows 11, 8 GB RAM, Intel i3 or better) and use the same machine for both the online and physical days. You also need a Gmail account and an active credit card, net banking or UPI method to activate paid Google AI services and a Gemini API key before the batch begins. ICAI notes you can remove the saved payment method and API key after the capstone.

Honest assessment of Level 2:

Strengths:

• Genuinely technical, and much more so than it used to be — Python, PWAs, APK builds, MCP connectors and local models are real skills, not an awareness tour
• It teaches the current stack, not a generic one — MCP, agentic patterns, n8n and local inference are what practitioners are actually adopting in 2026, and ICAI is teaching them roughly in step with the market rather than years behind it
• Not CA GPT-centric — this was the fair criticism of earlier versions and Ver 3.0 largely answers it. The syllabus leans on Claude, Gemini, LM Studio and third-party platforms
• Tally–Claude MCP is the standout module — a genuinely India-specific integration that no international AI course will teach you
• CPE value — 30 structured hours is an efficient chunk of the annual requirement
• Capstone forces production — you cannot pass by attending. You have to ship something

Weaknesses:

• The objective may not be your objective — this is a course about becoming an AI consultant to clients. If you want to use AI better in your own audit and tax work, you are paying for a build-and-deploy curriculum you may never use
• Five days is very little for this much ground — Python fundamentals, full-stack deployment and Android APKs occupy one day each. Nobody learns to build production software in a day per topic. Expect orientation, not competence
• Tool-specific and therefore perishable — Lovable, Antigravity, The Bricks, Comet, Opal. Several of these will look different or be gone in eighteen months, and the syllabus is pinned to them rather than to durable concepts
• Little audit-specific assurance content — for a certification aimed at CAs, there is strikingly little on SA compliance, documentation, or how AI-assisted work survives a peer review or NFRA inspection
• The paid-API prerequisite is a real cost and a real risk — participants activate billed Google AI services on a personal card. Budget for it, and remember to unwind it afterwards
• Rigid attendance rules — missing Day 1 cancels your registration. For partners in season, that is a genuine scheduling risk on ₹10,000

Who should do Level 2:

• ✓ CAs who genuinely want to build and deploy small AI tools — internally or for clients
• ✓ The designated "AI person" in a firm who needs to evaluate and assemble tools rather than just use them
• ✓ CAs advising SME clients on Tally-adjacent automation — the MCP module alone is close to worth the fee
• ✓ Anyone who needs 30 structured CPE hours and would rather earn them on something technical

Who should skip / supplement:

• ✗ CAs who want better AI-assisted audit and tax execution — that is not what this course is for. Level 1 plus practice will serve you better
• ✗ Practitioners already comfortable with Python, APIs and local models — you will spend three of five days on ground you have covered
• ✗ Partners who need firm-level AI strategy rather than hands-on build skills
• ✗ Anyone who cannot commit to all five days on the scheduled dates

AICA Level 3 — launched, but not yet a course you can take

This is where the public record contradicts itself, so it is worth being precise.

Level 3 was launched on 26 June 2026, at ICAI's AI Innovation Summit (AIS 2026) at Bharat Mandapam, New Delhi — a summit with 4,000+ delegates and 50+ speakers. Alongside it ICAI signed an MoU with Sarvam AI and released a practice guide, AI Essentials for Chartered Accountants. The Institute stated it had trained 50,000+ members in AI and built 150+ GPT-based tools. This was reported by The Tribune, Careers360, CAclubindia and TaxGuru, and described consistently across all of them as an advanced certification for members who have completed Level 2.

And yet: ICAI's own current Level 2 course document, Ver 3.0, still carries a "Future Roadmap" line reading "ICAI plan to Introduce AICA-Level-3 at later stage." As of late August 2026 — two months after the launch — there is no published Level 3 syllabus, no fee, no duration, no eligibility document and no batch calendar on ai.icai.org.

So the accurate statement is: Level 3 has been announced and formally unveiled, but it is not yet a course you can enrol in. Anyone telling you what Level 3 costs or covers is guessing.

What we can say with confidence:

• The prerequisite is AICA Level 2 — consistently reported, and consistent with the programme's structure
• Given Level 2 already reaches Python, app deployment and agentic AI, Level 3 has to go somewhere genuinely advanced to justify itself. The obvious candidates are firm-level AI strategy and governance, model evaluation and assurance over AI systems, or a consulting-practice track
• ICAI has historically published a detailed course document before opening batches, as it did for Level 2. That document, not a news report, is the thing to read before committing

The honest concern, which the launch has not resolved: is Level 3 substantive practitioner education, or a credential that sounds important without adding much beyond Level 2? With Level 2 now carrying real technical weight, the bar for Level 3 is higher than it was a year ago.

Practical advice: do not build a training plan around Level 3 yet. Plan for Level 1 and Level 2, and treat Level 3 as something to evaluate when the syllabus appears. If you want to be told the moment it does, the AI in ICAI helpline is ai@icai.in.

One footnote that says something about the programme's own relationship to the tools it teaches: ICAI's official "AICA Course Assistant" for participant queries is a custom GPT hosted on ChatGPT.

What AICA still doesn't cover (the gap)

This section needs an honest correction. Earlier versions of this post listed five gaps in the AICA syllabus. Ver 3.0 has closed two of them, and it is only fair to say so before listing what remains.

Now covered, and reasonably well:

• Agentic and multi-agent work — Day 1 and Day 5 cover agent architecture, MCP connectors and agentic use cases directly. The framework choices differ from the ones practitioners often reach for (ICAI teaches n8n and Google Opal rather than CrewAI, AutoGen or LangGraph), so the multi-agent post is still a useful complement — but this is no longer a gap
• Running models locally — the Offline GPTs module covers LM Studio, local execution and the privacy argument. What it does not cover is the money: what self-hosting actually costs and when it beats a subscription. The open-source LLM hosting post has the verified Indian numbers

Still genuinely uncovered:

1. Honest comparison of public LLMs

The syllabus teaches several models and platforms, but not how to choose between them for a given task. See the 4-LLM comparison post and which AI model CAs should use by task.

2. RAG implementation

How to build a Standards on Auditing or CARO knowledge base with retrieval — the single most useful architecture for an audit firm, and absent from the syllabus. See the RAG post.

3. Audit-tech vendor evaluation

How to assess an audit AI vendor. Understandably absent from an ICAI course, but it is what partners actually have to do. See the AI Audit Tool Evaluation Checklist.

4. Assurance over AI-assisted work

The biggest gap, and the one that matters most to a practising auditor. Nothing in the syllabus addresses how AI-assisted work is documented under SA 230, how it survives a peer review or NFRA inspection, or where the auditor's judgement must remain unaided. A certification aimed at CAs that teaches you to build and deploy AI tools, but not how to defend their output in a file review, has left the hardest part to you. See what AI can and cannot do in Indian audit today.

The pattern across all four: AICA now teaches you to build with AI competently. It still does not teach you to assure work done with AI — which is the part that is specific to being an auditor rather than a developer.

Decision framework — should you / your team take AICA?

For individual practitioners:

The key question for Level 2 is no longer "how deep does it go" but "do you want to build things?" Ver 3.0 is a build-and-deploy course. Answer that honestly and the rest follows.

For firms:

What AICA does well that gets understated

Despite the gaps, three things AICA does well:

1. Standardisation — when 1,000+ CAs take Level 1, AI literacy becomes baseline in the profession. This raises the floor for the whole industry.

2. ICAI branding — for client interactions, "our team has AICA certification" carries weight that "we use ChatGPT" does not. Marketing value is real.

3. Curated content — ICAI committees vetting material means lower risk of misleading content. For new entrants, this trust signal matters.

For these reasons, AICA Level 1 is a worthwhile foundation even if it's not sufficient for advanced practitioners.

Combining AICA with external resources

The optimal practitioner path:

There are now two sensible paths, and they diverge after Level 1.

If you want to use AI well in audit and tax work — the path most practitioners actually want:

1. AICA Level 1 (₹5K, 3 days) — get the foundation and the credential
2. This AI-in-audit series (free) — get the practitioner-level depth
3. Vendor tool training (often included with subscription) — get specific-tool proficiency
4. Self-directed practice — adopt tools in real engagements, learn through use

Total cost: ₹5K plus time. Level 2 is not on this path, because building apps is not what you are trying to get better at.

If you want to build and deploy AI tools — for your firm or for clients:

1. AICA Level 1 (₹5K, 3 days) — prerequisite anyway
2. AICA Level 2 (₹10K, 5 days) — Python, deployment, MCP, agents, and a capstone that forces you to ship
3. Sustained practice afterwards — five days is orientation. The capability comes from the next six months
4. The gaps this post lists — RAG, model choice, and how AI-assisted work survives a file review

Total cost: ₹15K plus considerably more time than the course itself.

Choosing the wrong one of these two is the most expensive mistake available here — not in rupees, but in five days of season time spent learning Android APK builds you will never use.

Bottom line

ICAI's AICA programme is more serious than its reputation suggests, and it has moved faster than most of the commentary about it. Level 1 remains an accessible, reasonably priced foundation. Level 2 at Ver 3.0 is a genuine technical course — Python, MCP connectors, local models, deployment, agents — aimed squarely at CAs who want to build. Level 3 has been launched but not yet published, and cannot be planned around.

The single most important thing to get right is that Level 1 and Level 2 are now aimed at different people. Level 1 makes you AI-literate. Level 2 makes you a builder. Most practitioners want the first and assume the second is simply more of it. It isn't.

For most practitioners:

• Take Level 1 if you have no prior AI exposure or want the ICAI brand on your CV
• Take Level 2 only if you intend to build — internal tools, client automation, Tally integrations. Not as a general CPE top-up
• Ignore Level 3 for now — there is no syllabus, fee or batch calendar to evaluate. Read the course document when ICAI publishes it

For firms:

• Sponsor juniors + managers for Level 1 as part of training budget
• Fund Level 2 for the one or two people who will actually build — not as broad staff development
• Build firm-level AI strategy outside AICA — using external resources, vendor partnerships, or consultants
• Cover the assurance gap yourself — AICA will not teach your team how AI-assisted work is documented and defended in a peer review

One alternative worth tracking: in August 2026 ICAI signed a letter of cooperation with Singapore's ISCA to explore extending ISCA's 30-hour AI Fluency Programme to Indian members. It is self-paced, case-driven and aimed at fluency rather than building — a closer fit than AICA Level 2 for CAs who simply want to use AI well. International enrolment exists through ISCA, but the ICAI-member route and India terms are not published yet. See what the ISCA tie-up actually changes.

For CAs who want a lower-commitment starting point before any course, ICAI's 2025-26 AI publication stack is now substantial enough to matter. Start with which ICAI AI publication to read first, then decide whether a certification is worth the time.

For firms turning that training into a service line, the Code of Ethics question is now live too: Can CAs offer AI consultancy? covers what changed in 2026 and where the audit-client independence boundary sits.

The series and content on CORAA's University (AI Lab, Prompt Library, Tool Guides, AI Evaluation Checklist) complement AICA by going deeper on specific tools and providing comparison content AICA doesn't.

The two together — AICA + practitioner-level external content — produce CAs who are credibly AI-literate and operationally proficient. Either alone is insufficient.

Try CORAA → Tools that put AICA learning into operational use. India-hosted audit AI, partner dashboards, full-population testing. See pricing · AI Lab · AI Audit Tool Evaluation Checklist.

Statutory facts on this page are checked against their sources, and the page says where it relied on secondary reporting. How we verify · Report an error