Compliance is painful, bureaucratic, and often paper-based, so it has long persisted as being manual and human intensive.

That friction has historically made compliance a graveyard for startups.

But AI may finally go from "good enough to pilot" to "good enough to trust".

In legal, broad model choice and consistently high accuracy gave teams the confidence to finally embrace AI. Many LLMs now score 80-100% on LegalBench’s 162 legal reasoning tasks.

This matters directly for compliance, because compliance is essentially applied legal reasoning under operational constraints, built on the same core tasks: reading regulatory text, applying rules to fact patterns, identifying exceptions, and flagging ambiguities.

Full piece from a16z's @jamdac and @astrange: https://t.co/niRB3jPioN

likes: 97 | retweets: 14 | replies: 21 | views: 24649