Goldman Sachs Partners With Anthropic to Automate Accounting With Claude

Goldman Sachs partners with Anthropic to build autonomous AI agents that will automate core accounting and compliance functions across the bank. The deal, revealed on February 6 by Goldman’s chief information officer Marco Argenti in an exclusive CNBC interview, represents one of the largest enterprise AI deployments in Wall Street history and marks a decisive shift in how the financial industry approaches back-office automation.

For the past six months, Anthropic engineers have been physically embedded inside Goldman’s technology teams, co-developing agents built on the company’s Claude model. The agents target two specific areas: trade and transaction accounting, where reconciliation and error resolution have long been slow and resource-intensive, and client vetting and onboarding, a compliance-heavy process involving enormous volumes of documents and strict regulatory checks. Argenti told CNBC the firm expects to launch the agents “soon,” though he declined to provide a specific date.

“Think of it as a digital co-worker for many of the professions within the firm that are scaled, are complex and very process intensive,” he said.

Goldman Sachs partners with Anthropic after an unexpected discovery. The bank began its AI push last year by piloting Devin, an autonomous coding agent built by startup Cognition, which is now broadly available to Goldman’s thousands of software engineers. When the firm tested whether Claude’s reasoning abilities could extend beyond writing code into other complex, rules-based work, executives were caught off guard by the results. Argenti told CNBC the firm was “surprised” at how capable Claude was in areas like accounting and compliance that combine the need to parse massive data sets while applying regulatory judgment. “Is that because coding is kind of special,” he said, “or is it about the model’s ability to reason through complex problems, step by step, applying logic?” The answer turned out to be the latter, and it changed how Goldman thinks about the scope of what AI can do inside a bank.

The strategic backdrop makes the timing significant. Goldman Sachs partners with Anthropic against the context of a sweeping internal restructuring that CEO David Solomon launched in October 2025 under the banner of “OneGS 3.0.” In a firmwide memo signed by Solomon, President John Waldron, and CFO Denis Coleman, the bank called for a multiyear overhaul of its operating model built around AI. “The rapidly accelerating advancements in AI can unlock significant productivity gains for us,” the memo stated. Goldman disclosed a “limited reduction in roles” and pledged to constrain headcount growth through the end of 2025. The firm reported $15.18 billion in revenue and $4.1 billion in net income in the third quarter of 2025, with investment banking fees surging 42% year over year to $2.66 billion, meaning it is making this bet from a position of financial strength rather than desperation.

Solomon, speaking publicly in the weeks that followed, framed the AI transition as analogous to past technology revolutions but acknowledged one critical difference. “The pace is so quick,” he told CNBC, “I think there’s a possibility that there’s a little bit more volatility, or an unsettled transition around certain job functions.”

The choice of partner is telling. Goldman Sachs partners with Anthropic rather than rivals like OpenAI or Google DeepMind in large part because Anthropic has built its reputation around safety, interpretability, and reliability. In a regulated financial environment where compliance errors can trigger billions of dollars in fines and lasting reputational damage, those qualities carry particular weight. Claude’s ability to process long documents, maintain context across multi-step workflows, and provide traceable reasoning appears to have been a deciding factor. Goldman already runs multiple AI models internally through its GS AI Platform, which integrates tools from OpenAI, Google, and Meta behind the firm’s firewall. But for the sensitive, high-stakes work of automating accounting and compliance, the bank landed on Claude.

The agents are being designed to review millions of transactions annually, match records, flag discrepancies in trade reconciliation, and navigate the dense regulatory frameworks governing know-your-customer and anti-money laundering protocols. Goldman employs thousands of people in the compliance and accounting functions where these agents will operate. Argenti told CNBC it was “premature” to expect the technology would lead to job losses for those workers. He was more direct about another consequence: Goldman could eventually cut out the third-party software providers it relies on today. “It’s always a trade-off,” he said.

That remark sent a chill through the enterprise software sector. Goldman Sachs partners with Anthropic at a moment when investors are already reassessing which companies will survive the AI transition. The announcement landed in the same week that Anthropic’s latest model updates contributed to a sharp selloff across software stocks, as traders wagered that autonomous agents could displace the specialized compliance, accounting, and workflow platforms banks have spent years and billions licensing. If Claude can handle trade reconciliation and regulatory monitoring at scale, the commercial case for maintaining those expensive third-party tools starts to erode.

Goldman Sachs partners with Anthropic as the bank’s broader AI ambitions continue to pick up speed. Beyond accounting and compliance, Argenti told CNBC the firm could next develop agents for creating investment banking pitchbooks and conducting employee surveillance. The bank’s GS AI Assistant, an internal ChatGPT-style tool, was opened to all 46,500 employees in June 2025 and has already surpassed 50% adoption, with a target of reaching universal usage by the end of this year. Solomon himself has noted that AI can now draft 95% of an S-1 filing, the document a company submits to the SEC when going public, in minutes. That task used to take a six-person team two weeks. Of Goldman’s more than 47,000 employees, about 13,000 are engineers, a ratio that has held at around a quarter of the workforce for years but could shift meaningfully as agents absorb a greater share of the workload.

The competitive pressure across Wall Street is intense. JPMorgan Chase, Morgan Stanley, and other major banks are each pursuing their own AI strategies. But the Goldman and Anthropic collaboration stands out for its scope, its depth of integration, and the candor with which its architects have discussed what comes next. The bank has long relied on armies of junior accountants, compliance officers, and analysts to handle the procedural, document-heavy work that keeps a global institution operating within the rules. The question it is now testing, with Anthropic engineers sitting inside its offices writing the software, is whether machines that reason like coders can do that work better, faster, and more reliably than the people who have done it for decades.

Wall Street, as Argenti put it, is finding out that it is always a trade-off.