Goldman Sachs restricted Anthropic's Claude Code and Microsoft's GitHub Copilot to its approximately 12,000 engineers in October, revoking access for traders and bankers who had used the tools to monitor client flows, model market positions, and generate materials. The policy follows an April removal of Claude for Hong Kong bankers due to contractual and jurisdictional constraints, and reflects a risk review citing productivity concerns, uncontrolled code execution on machines handling sensitive data, and information barrier violations between banking and trading groups. Chief Information Officer Marco Argenti had promoted "vibe coding" by non-technical staff as recently as July, describing a workflow where traders built prototypes for engineering refinement. However, the bank determined that agent-written code deployed without formal review, testing, or deployment controls posed unacceptable operational and regulatory risk, particularly when authored by personnel unable to verify edge-case behavior. Cost factors, including token-based billing, may have contributed. Goldman continues to deploy Claude-based agents for trade accounting and client onboarding through engineering teams and serves as lead underwriter for Anthropic's upcoming Nasdaq IPO (ticker: ANT). Barclays, by contrast, targets 50% Claude Code adoption among developers while cutting 5,000 roles. JPMorgan and Schwab have also restricted external agents from customer accounts. The article argues Goldman's restriction is justified but poorly executed, recommending clear ownership of agent-generated code, mandatory engineering sign-off before production use, and systematic capture of displaced workflows.