Is Claude Coworker welcome in your finance department?

Your finance team is ticking along nicely but someone has invited Claude Coworker in. No IT sign-off no security review, no one asked you. Claude Coworker can access your files, run tasks, move through your systems. Autonomously for £20 a month your team can do this without you knowing.

Is that a bad thing?

Not necessarily. Claude Coworker is genuinely impressive. But impressive and governed are two very different things. Is this causing you sleepless nights? It would me in my CFO days. The question is not whether AI belongs in your finance team. It does. The question is whether it arrives before your guardrails do.

Finance has always had a built-in truth mechanism Think about how accounting works at its core. Every entry has an opposite. Every debit has a credit. If something is deleted, the trail remains. If a number moves, something else moves with it. The system is self auditing by design. That's not just good practice. It's the architecture of trust that finance is built on.

We always know who did what. We know when they did it. We know what changed and what the original figure was. Accountability isn't an add-on in finance. It's structural. "In finance, we always have someone accountable. We know who did what. There's always an audit trail." Now ask yourself. Is that still true when Claude Coworker is involved?

Claude Coworker doesn't work like a ledger When Claude Coworker runs a task, builds a model, or produces an analysis, there's no opposite entry. There's no audit trail showing you what files were accessed, what data was used, what assumptions were made, or what version of the output was shared.

If the numbers look right, they look right. But looking right and being right are not the same thing. Here's what concerns me most. Claude Coworker can produce an analysis that balances. That looks clean. That passes a sense-check. And it might still be wrong. Or it might be built on data that should never have left your systems.

The error won't announce itself. That's what makes it different from a manual mistake. "Claude Coworker can make a spreadsheet or analysis look like it balances. But that's not the same as it being right."

The five questions you need to answer before you say yes I'm not suggesting you lock Claude Coworker out of your finance team. The efficiency gains are real. The capability uplift is real. But as the CFO or senior finance leader, these are the questions that need answering. And right now, most teams don't have the answers.

1. What is Claude Coworker actually touching? When your team enables Claude Coworker, which files can it see? Which systems can it access? Do you know? Has anyone checked?

2. Who owns the output? If Claude Coworker produces a financial analysis and it goes into a board pack, who is accountable for it? Your team member? You? The AI? This isn't a philosophical question. It's a governance one.

3. Do you have an AI policy in place? Not a technology policy. Not a general data policy. A specific AI policy that covers what tools are permitted, what data can be accessed, what outputs need human review, and what the escalation path looks like if something goes wrong.

Here is something most people do not know. Anthropic themselves advise against granting Claude Coworker access to financial documents. That is not a third party warning. That is the people who built it. And Cowork activity is currently not captured in audit logs. Anthropic are explicit. Do not use Cowork for regulated workloads. Finance is a regulated workload. Source: Anthropic Help Centre — Use Claude Cowork Safely ttps://support.claude.com/en/articles/13364135-use-claude-cowork-safely

4. Have you trained your team? Not just on how to use it, but on how to check it? Using Claude Coworker well in finance isn't just about knowing what to ask it. It's about knowing when to trust the output and when to challenge it. That's a skill. It needs to be built deliberately.

5. What's your audit trail? If your auditors ask how a number was derived, and the answer is Claude Coworker produced it, what happens next? Have you thought about that conversation? What good looks like

The good news is that this is solvable. Finance teams that are getting this right aren't blocking Claude Coworker. They're building the infrastructure around it. An AI policy that's actually been read and signed off. Clear guidance on what Claude Coworker can access and what stays off limits. A review process for outputs before they go into anything material. And a culture where it's safe to say: I used Claude Coworker for this. Here's how I checked it.

The CFOs who get this right aren't the ones who move fastest. They're the ones who can justify the output. "AI in your finance team isn't the risk. Having it arrive before your guardrails do. That's the risk."

One question worth sitting with If your auditors walked in tomorrow and asked to see your AI governance framework, what would you show them? If the answer is nothing. That's not a reason to panic. It's a reason to start.

The finance function has always been the place where rigour lives. That doesn't change because the tools do. If anything, it matters more than ever.