NEW YORK, Feb 8, 2026, 08:25 EST
Goldman Sachs is working with AI startup Anthropic to build “AI agents” that automate internal banking work such as trade and transaction accounting and client due diligence, the checks banks run to vet customers, CNBC reported on Friday. Goldman confirmed the report’s accuracy, Reuters reported. (Reuters)
The move matters because it pushes beyond chat-style assistants into software that can take multi-step actions on its own — the kind of automation banks have been cautious about in heavily controlled areas. The agents are being tested on transaction reconciliation, trade accounting, client vetting and onboarding, tasks that have “resisted automation for decades” because they must be run against strict regulatory frameworks, PYMNTS reported. (Pymnts)
It also lands in a touchy market moment for anything that looks like “AI that does jobs.” Reuters reported that a selloff in software and data-and-analytics firms was triggered by a new plug-in from Anthropic’s Claude, and strategist Carlota Estragues Lopez said, “Headlines that would have pushed shares to fresh highs during the peak of AI optimism are now being interpreted far more cautiously by investors.” (Reuters)
Goldman’s Chief Information Officer Marco Argenti told CNBC the firm is “in the early stages” and expects to launch agents “soon,” without giving a date, Sharecast reported. He called talk of job losses “premature,” but said the bank could cut out some third-party providers as the technology matures. (Sharecast)
Argenti pitched the idea as speed, not splash. “Think of it as a digital co-worker for many of the professions within the firm that are scaled, complex and very process-intensive,” he said, according to Observer, which also reported Argenti said similar agents could later be used for employee surveillance and to draft investment-banking pitchbooks. Goldman and Anthropic did not respond to Observer’s requests for comment. (Observer)
Outside Goldman, banks are trying to work out what to automate first — and how to prove it pays off. Alexandra Mousavizadeh, cofounder and co-CEO of Evident, told Business Insider that some AI capabilities are now “table stakes,” while the use of AI agents is still early, especially in externally facing roles where humans are likely to stay in the loop as banks decide on guardrails. (Businessinsider)
Goldman has been moving in this direction for a while. A Reuters report in June 2025 said the bank launched its GS AI Assistant firmwide, with about 10,000 employees already using it, according to an internal memo sent by Argenti. (Reuters)
Rivals have been building internal platforms and workflow tools that point in the same direction, even if they describe them differently. Citi said in September 2025 that its proprietary Citi Stylus Workspaces was upgraded with “Agentic AI” and integrated with select internal systems; Morgan Stanley said in June 2024 it launched an OpenAI-powered “Debrief” tool that creates meeting notes and drafts follow-up messages with client consent; and JPMorgan said in 2025 that its LLM Suite roadmap includes building AI agents that can carry out a series of actions to reach a goal. (Citigroup)
But the downsides are not theoretical. Britain’s Financial Conduct Authority warned in December that “agentic AI” introduces new risks because of the ability for something to be done “at pace,” and said autonomy and speed can magnify governance and stability risks when multiple agents interact. (Reuters)
Goldman has not put a public date on when its agents go live, and the first real test will be boring in the way regulators prefer: audit trails, controls, and whether the software stays reliable when the work gets messy. If the agents can’t hold up under edge cases, they won’t save time — they’ll just move the bottleneck to cleanup.
