OpenAI and Anthropic intensify battle to power AI in finance

What’s notable in this wave of coverage isn’t that Anthropic and OpenAI are selling into finance—that’s expected—but how quickly they’re moving from generic chatbots to workflow‑native agents with deep data access and governance hooks. Anthropic’s ten pre‑built Claude agents for KYC, pitchbooks, and month‑end close, paired with the FIS Financial Crimes AI Agent, are designed to sit inside regulated workflows rather than around them. OpenAI’s PwC alliance and private‑equity joint ventures aim to do the same for revenue recognition, procurement, and back‑office finance.

This is strategically pivotal because it turns frontier models into infrastructure, not just tools. Once an AI agent is embedded in AML investigations or audit prep, ripping it out is costly; that stickiness supports the massive compute contracts these labs are signing. It also pushes both firms to specialize models and agents around domain benchmarks, compliance requirements, and integration patterns instead of just leaderboard scores.

For the race to AGI, the financial sector is becoming a live‑fire test bed for agentic systems that act, not just assist, under tight oversight. Success here will harden patterns—like chain‑of‑command, logging, and constrained autonomy—that will likely be reused in more open‑ended domains. It also raises the stakes: if AI can safely reason about trillions of dollars in capital flows, expectations for what counts as “general” intelligence will ratchet higher.