How AI Is Changing Compliance and Audit Work

This post summarizes “Compliance and Audit”, a Y Combinator request-for-startups video published on the Y Combinator channel in January 2025. The video says almost 4 million people in the US and Europe work in compliance and audit (a figure we have not independently checked), and that compliance costs are only rising. It lists regulations from GDPR and Dodd-Frank to AML, KYC and ESG reporting. Traditional compliance tasks are tedious and time-consuming, and the video argues that AI and automation can help.

Key Takeaways #

• Compliance costs are rising as the regulatory environment keeps expanding.
• Traditional tasks are manual and repetitive.
• The video argues LLMs can automate much of the testing auditors do by hand.
• A well-trained model could review every document instead of a sample.

The Growing Need for Compliance #

Companies have to keep up with evolving regulations. Compliance professionals read dense regulations, cross-check internal policy and procedure documents, and produce repetitive reports.

The Challenges of Traditional Compliance Tasks #

Per the video, traditional compliance work involves:

• Reading regulations: dense regulations have to be read and understood.
• Cross-checking policies: internal policy and procedure documents are checked against them manually.
• Sampling work: frontline work is sampled manually, so issues in unsampled work can be missed (our inference).
• Repetitive reporting: reports are repetitive to produce. Auditors also sift through large amounts of unstructured data to detect issues.

These workflows, the video says, beg for automation.

How AI Can Help #

The video says LLMs excel at this work. It names four uses:

1. Reviewing documents: LLMs can parse regulatory documents, corporate policies and financial statements and highlight issues for human review.
2. Spotting anomalies in data.
3. Identifying incomplete records.
4. Highlighting contradictory policies.

Instead of sampling a few documents, the video says, a well-trained model can look at everything at once. It ends with the aim of continuously auditing every company, and invites founders building in this space to get in touch. That is a goal, not something the video shows working.

Our Take #

These are Y Combinator’s claims, and we have not tested them. If they hold, compliance professionals would spend less time on paperwork and more on judgment calls. Any model that flags issues still needs a human to review what it finds.