Goldman Sachs is no longer just a Wall Street titan—it's a bellwether for the future of AI in finance. The bank's recent partnership with Anthropic to automate compliance functions using AI agents marks a seismic shift in how enterprises approach regulated workflows. But as the dust settles on this partnership, a deeper question looms: Who governs the AI?
For six months, Goldman Sachs embedded Anthropic engineers to co-develop AI agents tailored for two critical areas: trade/transaction accounting and client vetting/onboarding. The results? A system built on Anthropic's Claude model, which the bank was "surprised" by its ability to handle both rule-based tasks and judgment-heavy decisions.
This isn't just another pilot. Goldman already deployed Devin, Anthropic's autonomous coding agent, to all its engineers. Now, the bank is scaling AI into compliance, a move CEO David Solomon has framed as part of a multiyear plan to reorganize around generative AI. "We'll constrain headcount growth," he warned, signaling a shift toward AI-driven efficiency.
The next targets? Employee surveillance and investment banking pitchbooks. As Goldman's CIO put it:
"Think of it as a digital co-worker for many professions that are scaled, complex and process-intensive."
Goldman's move isn't an outlier. A 70%+ adoption rate of agentic AI in banking (per EY) shows the sector is racing to automate. Yet the reality is stark: 16% of firms are fully deployed, while 52% are running pilots—with no robust governance frameworks in sight.
Regulators are equally split. The U.S. is deregulating, the EU is simplifying, and the UK is prioritizing growth. But with CCPA Automated Decision-Making Technology regulations set to take effect in 2026, and the SEC now labeling AI as a "clear area of operational risk" linked to cybersecurity, the pressure is mounting.
For compliance professionals, this is a wake-up call. AI isn't just a tool—it's a standing board agenda item.
Goldman's success with Claude highlights a critical challenge: training data provenance. Where did the model learn? Could it inadvertently encode IP violations, bias, or discrimination?
Then there's opacity. In regulated environments, an AI that can't explain its decisions isn't just unmanageable—it's unacceptable. And what happens if Anthropic, or any AI vendor, faces financial instability? The compliance fallout could be catastrophic.
The stakes are clear: audit trails, explainability, and third-party risk controls are non-negotiable. Yet few enterprises have the expertise to navigate this terrain.
Goldman's partnership with Anthropic legitimizes AI compliance agents for the entire industry. But here's the rub: Who audits Claude's compliance decisions?
Enter the new frontier of AI governance. Enterprises need experts who understand both security and AI—individuals who can design frameworks that balance innovation with accountability. This isn't just about ticking boxes; it's about building trust in systems that make high-stakes decisions.
The good news? The tools exist. From model risk management to explainable AI (XAI), the cybersecurity and compliance sectors are beginning to align. But the human expertise to oversee these systems? That's still scarce.
With CCPA 2026 on the horizon and the SEC's scrutiny intensifying, organizations can't afford to wait. The AI governance framework must be built now—before the next regulatory wave hits.
The future of AI in finance is here. But governance? That's up to you.