New AI Safety Company Raises $10 Million to Keep Bank Messages Compliant

A startup called ZeroDrift just secured $10 million in funding to solve a growing problem: how to make sure that when banks and insurance companies use AI to send messages to customers, those messages don't break any rules.

The funding round was led by Andreessen Horowitz, a major investment firm. ZeroDrift has already found customers — some of the country's largest banks, investment firms, and insurance companies have started using the system. That matters because if these companies send something that violates financial regulations, the penalties can cost hundreds of millions of dollars.

How ZeroDrift Works

Think of ZeroDrift as a security guard standing between an AI system and the outside world. Before an AI-generated email, text message, or video call leaves a bank's building, ZeroDrift checks it. Is it saying something that breaks a banking rule? Does it violate privacy laws? Only if it passes the check does the message actually go out.

The key difference from the way compliance works today is timing. Usually, banks check communications after they've already been sent — and then try to fix problems after the fact. ZeroDrift checks everything in real time, before it goes out. That's harder to do technically, especially because financial messages need to move very fast. A customer expects a response in less than one-tenth of a second.

ZeroDrift also has to handle different types of communication. A text message needs different checking rules than a video call. And the same message might be legal in the United States but illegal under European privacy rules.

Why This Matters Now

Banks are under pressure from two directions at once. On one hand, regulators around the world are tightening rules about what AI can and cannot do. The European Union started enforcing its AI rules in 2024, and U.S. banking regulators have made it clear that banks are responsible for everything their AI systems send to customers.

On the other hand, banks want to use AI to compete and serve customers faster. The problem is, right now they're scared to do it. Without a way to check AI messages automatically before they go out, it's too risky.

ZeroDrift's pitch is straightforward: with us, you can deploy AI safely. We'll catch the problems before they happen.

The Competition

Large companies like Thomson Reuters and Bloomberg have tools that check whether communications follow the rules, but those tools were built for human employees, not AI. An AI can generate thousands of messages a minute, and those tools can't keep up.

Some newer companies have started building AI oversight tools, but most focus on other problems — like figuring out if an AI model is biased or misbehaving. ZeroDrift is specifically targeting the one problem that costs banks the most money: messages that break the law.

What Could Go Wrong

Putting a checkpoint between an AI system and the outside world creates its own risks. If ZeroDrift's system fails or slows down, an entire bank's customer communications could get stuck. Banks would need to trust that ZeroDrift's technology is reliable.

There's also a practical problem: most big banks run very old, complicated systems that don't easily connect to new tools. ZeroDrift will need to integrate with dozens of different systems, and each one might have its own quirks. For security reasons, banks often want this kind of software running on their own computers, not on the cloud — which makes it harder for ZeroDrift to process information quickly at scale.

We have seen similar challenges before. When firewall software first came out, security companies struggled to block bad traffic without slowing down good traffic. The companies that succeeded started with one specific problem they could solve very well, rather than trying to solve everything at once.

What's Next

The $10 million funding level tells you what investors think about the market size right now. They see a real business, but not necessarily a huge one yet. The fact that Andreessen Horowitz is involved is worth noting — they have invested in many AI companies, so they may be thinking about how ZeroDrift could grow into other industries like healthcare and law, where AI compliance matters just as much.

ZeroDrift's biggest risk is speed. Banks are so worried about the compliance problem that they might just add AI compliance features to the software they already use — tools made by much larger companies. ZeroDrift would need to prove that its specialized approach works better before the big players catch up.

The regulatory rules around AI are still changing. New requirements appear every few months in different countries. ZeroDrift's system has to stay flexible enough to adapt to new rules without slowing down. That's a real engineering challenge.

If ZeroDrift succeeds, it could prove that real-time compliance checking is something every company needs when they use AI. That could expand the market beyond banking to healthcare, law firms, and other industries where sending the wrong message can have serious legal consequences.