Goldman Takes Back Traders' AI Coding Tools: the Experiment Is Over, Engineers Only
Bloomberg reported on 2026-11-02 that Goldman's October internal guidance pulled Claude Code and GitHub coding tools back to engineers only, with traders and banking staff losing access and the months-long vibe coding experiment over; Goldman says the tiering follows an assessment of productivity and risk management.
- 01THE RELEASE — Bloomberg reports on 2026-11-02 on Goldman's October internal guidance, pulling coding tools back to engineers only.
- 02THE CONFIRMATION — PYMNTS and SiliconReport independently relay and confirm the scope, background and Goldman spokesperson's statements the same day.
- 03THE WATCHPOINT — Whether other big banks follow with tiering, and whether business teams shift to compliant alternatives.
WHAT HAPPENEDWhat Happened
Bloomberg reported on 2026-11-02 that Goldman Sachs updated its internal guidance in October, pulling access to Anthropic's Claude Code and Microsoft's GitHub coding tools back to software engineers; traders and investment banking staff lost access, ending an experiment of several months in which the tools were open to everyone.
Use during the experiment had already gone well beyond writing code: traders and bankers used the tools to monitor client trading flows, simulate the impact of market swings on positions, and draft text and chart materials; according to people familiar with the matter, the withdrawal of access has caused discontent among some traders.
The scale of the reversal is worth noting: in July 2026, Goldman's chief information officer Marco Argenti was publicly praising non-technical staff's vibe coding, saying they built prototypes with AI tools and handed them to the firm's roughly 12,000 engineers to polish and ship; four months later, the same group was excluded from the coding tools.
Spokesperson Nick Carcaterra's explanation was procedural: open the tools for experimentation first to find the best use cases, then evaluate on factors including productivity and risk management, and finally give the tools to those best suited to long-term use. No public incident, cost figure or internal audit conclusion supports the tightening.
The tightening has boundaries and is not a blanket ban on AI: engineers keep the coding tools, and strats roles, which span trading, quantitative research and software engineering, are reportedly not covered by the restriction; the GS AI Assistant, rolled out firmwide in June 2025, and internal general-purpose assistants remain in operation.
WHY IT MATTERSThe End-of-Experiment Signal Matters More Than the Ban Itself
Goldman's wording is deliberately mild: it does not say the tools are bad, or that staff misused them, only that the experiment period is over and the user population is being narrowed by productivity and risk assessment. That is exactly the standard second phase of AI adoption at large institutions: open up to find use cases first, then tier access by role and risk.
For other financial institutions, Goldman's version is a reusable template: not an outright ban, but the strongest coding tools restricted to those best able to handle them, while general-purpose assistants stay open to everyone. The next signal is whether other big banks adopt the same tiering, or keep opening up.
WHY IT MATTERSWall Street Is Tightening in the Same Direction, Just in Different Ways
Goldman is not an isolated case: JPMorgan CEO Jamie Dimon warned about rogue agents on the October earnings call, saying external AI agents would not be allowed into the bank; Bank of America then limited agent access to consumer deposits; and Citi banned third-party agents from logging into client accounts or trading.
The four institutions' moves together point to the same judgment: the value of AI tools no longer needs proving; the boundaries do. Whoever writes role tiering, agent permissions and accountability down first pays less tuition in the next round of AI deployment.
SIGNAL JUDGMENT — ANALYSISGoldman Draws a Role Boundary Around AI Tools
Tokenta Signal's judgment: the real signal in this tightening is not that Goldman has turned against AI coding tools, but that the all-staff experiment phase is over. AI deployment at large institutions is moving from anyone can use it to tiered access by role and by risk. The next signal is whether other big banks copy Goldman's tiering template, and whether the business teams that lost access shift to compliant low-code alternatives.
WHAT REMAINS UNCERTAINWhat Remains Uncertain
EVIDENCEEvidence
Bloomberg's reporting was independently relayed by PYMNTS and SiliconReport and confirmed by two sources; the original link was opened and verified.
PRIMARY SOURCESPrimary Sources
- PYMNTS relay of Bloomberg reporting
https://www.pymnts.com/news/artificial-intelligence/2026/goldman-sachs-bars-bankers-from-employing-some-ai-tools/ - SiliconReport report
https://www.siliconreport.com/goldman-sachs-cuts-off-most-staff-from-claude-code-and-github
RELATEDRelated Intelligence
- Microsoft Winds Down Its Oracle AI Cloud Rental as Older Nvidia Chips Fall Behind
- An Unreleased Anthropic Model Proved an 18-Year-Old Quantum Conjecture, Machine-Checked in Lean
- IBM Unveils Quantum System Two in Chicago, Its Most Advanced Computer Yet, Powered by Nighthawk
- Vistra Posts $1.14B Q3 Profit as Power EBITDA Jumps Over 50%, Lifts 2026 Outlook