Equals Money lets customers’ AI tools read data but not move money

Equals Money is opening up to customer AI agents, but only so far. The payments company is exposing a Model Context Protocol server to let those tools pull data faster, while keeping a firm lock on anything that would let an agent move money. That’s the dividing line in fintech right now: read access is useful, write access is where the trouble starts.

James Simcox, Equals Money’s chief operations and product officer, said the company already uses agents in customer-facing roles and estimates that AI now writes about 92% of its code. With that level of use, he wants agents treated as more than just software helpers. They need their own identity, their own access, and a way to be shut down quickly if they go off script.

Okta has been pushing in that direction too, with runtime enforcement and a broader kill switch for AI agents. Equals can already cancel agent tokens, but Simcox said the process is manual. What he wants instead is an automated system that spots bad behavior and cuts everything off before it spreads. His point was blunt: if one human goes rogue, that’s one problem. If one agent goes rogue, it can turn into 100,000 at once.

The company is applying the same caution to customer-side tools. Simcox said Equals behaves as though it carries full liability for customer errors and checks each flow multiple times. So even if a customer connects an enterprise AI tool to the MCP server, the deal is limited. Data can be read. Some non-sensitive writing is allowed. Payments are not.

That may sound conservative, but in regulated finance conservative is often just another word for survivable. Once an agent can act, every shortcut becomes a liability question with a very expensive answer.