Anthropic launches finance AI agents targeting Goldman Sachs work

Anthropic is aiming squarely at the kind of bank work that eats junior hours: pitchbooks, KYC screening, financial statement review, month-end close, and the first drafts that sit behind client prep and credit work. On May 5, the company unveiled 10 ready-to-run finance agent templates and said they were built to plug into the tools bankers already use, including Excel, PowerPoint, Word and, soon, Outlook. It also said Claude Opus 4.7 was the best fit for the stack, citing a 64.37% score on Vals AI’s Finance Agent benchmark.

For Goldman Sachs employees, the immediate message is not that AI is replacing bankers overnight. It is that the baseline for production speed is being reset. Tasks that once justified late nights, associate overnights and endless markup cycles can now be accelerated by software that is designed to draft, revise and move across documents without losing context. That shifts routine work into a faster lane, while pushing bankers toward the parts of the job that still require human judgment: client positioning, deal nuance, risk calls, and the polish that turns a competent draft into something a managing director will send.

The launch lands at a sensitive moment for Goldman. On April 28, Reuters reported that Goldman removed access to Anthropic’s Claude for bankers in Hong Kong, while Gemini and ChatGPT remained available on the firm’s internal platform. Anthropic’s spokesperson said Claude had never been officially supported in Hong Kong. That split captures the central tension for banks now: the same tools that promise speed also raise questions about data control, cyber risk and where sensitive material can safely flow.

Goldman is not standing outside the market, either. On May 4, Anthropic announced a new enterprise AI services company with Goldman Sachs, Blackstone and Hellman & Friedman, backed by General Atlantic, Leonard Green, Apollo Global Management, GIC and Sequoia Capital. Goldman chief information officer Marco Argenti has already framed AI as more than chatbots, saying in January that 2026 would be an even bigger year for change. Goldman Sachs Research has also estimated that 300 million jobs globally are exposed to automation by AI.

That is why Anthropic’s move matters inside Goldman as much as it does in the market. The work most likely to get faster is the mechanical middle of banking: document drafting, statement review, KYC, and the assembly of pitch materials. The work most likely to be redesigned is anything that can be standardized into repeatable workflows across Microsoft apps and approved templates. The work that should remain harder to automate is the most sensitive, judgment-heavy layer, where client trust, compliance decisions and commercial judgment still decide who wins the mandate.