Goldman Sachs teams up with Anthropic to automate accounting and compliance roles

After six months of stealth development, Wall Street giant deploys autonomous agents for high-stakes accounting and compliance.

Editorial credit: ioda / Shutterstock

Goldman Sachs is moving beyond simple AI experimentation and embedding "digital co-workers" into the heart of its most sensitive operations. On February 6, 2026, the bank’s Chief Information Officer, Marco Argenti, revealed that Goldman has been working with Anthropic for six months to co-develop autonomous agents powered by Claude Opus 4.6 to handle trade accounting and regulatory compliance.

The initiative represents a fundamental shift in banking - using AI not just to write code or answer emails, but to manage a portion of the bank’s $2.5 trillion in assets under supervision.

From coding to "cognitive" finance

The pivot to Anthropic began when Goldman’s tech team noticed a surprising trend. While testing Claude for software engineering (building on their use of the "Devin" AI coder), they realized the model’s step-by-step reasoning was perfectly suited for the rigid, logic-based world of finance.

The bank processes millions of transactions annually. The new Claude-powered agents can independently review trades, match internal records against external bank statements, and flag discrepancies in real-time, a task that previously required armies of back-office staff.

Client onboarding (KYC/AML)

Onboarding a new institutional client is notoriously slow due to "Know Your Customer" and "Anti-Money Laundering" regulations. Goldman reports that using Claude to parse global databases and legal documents has already slashed onboarding times by 30%.

Argenti told CNBC that the bank was "surprised" by Claude's ability to apply professional judgment to complex datasets. "Is that because coding is special, or is it about the model’s ability to reason through complex problems?" he asked, answering that it was clearly the latter.

The workforce reality: "Digital colleagues"

The bank is currently framing this as an augmentation strategy rather than a replacement one. However, the long-term implications for hiring are significant.

To start with, CEO David Solomon has previously stated that generative AI is central to the organization’s multi-year plan to "constrain headcount growth" while increasing revenue.

Also, while internal job loss concerns were labeled "premature," Argenti admitted that mature AI agents could allow Goldman to cut ties with expensive third-party service providers that currently handle overflow accounting and compliance work.

New use cases

The bank is already thinking about the next phase, which includes using AI for employee surveillance (monitoring for insider trading) and creating high-stakes investment banking pitchbooks.

Market impact: A warning for SaaS

The news of Goldman building its own "agentic layer" using Anthropic sent shockwaves through the enterprise software market.

Investors are increasingly worried that if global banks can build custom autonomous systems, they will stop paying for legacy "Software-as-a-Service" (SaaS) licenses. Consequently, the "middleman" vendors of the financial world may find themselves obsolete by the end of the decade.