The 10-Second Version (AI Summary)
What happened: Goldman Sachs limited Claude Code and GitHub coding tools to its roughly 12,000 engineers, cutting off bankers and traders.
Why it matters: Big banks are drawing hard lines on AI access, and dealerships handle the same kind of sensitive customer data.
What to do: Set and communicate your store's AI policy before a breach or a cutoff forces it.
AI-generated and reviewed by the ASOTU team.
What breaks first when your AI tools disappear?
We still remember the last long ChatGPT outage, when LinkedIn went quiet for a full day. That's how fast these tools become part of how people work. This week, some Goldman Sachs traders got a taste of the same thing on purpose.
On today's episode, Paul made the point that you don't need a year with these tools for them to change your work. Ninety days in, they've already changed how much you get done and how you think through problems. Losing them overnight wrecks your flow.
Goldman Sachs is limiting access to Anthropic's Claude Code and Microsoft's GitHub coding tools to its engineers.
Updated internal guidance from October came to light Monday. Access is now set by job type instead of location. The bank's roughly 12,000 engineers keep the tools, and its quantitative strategists, or "strats," are exempt. Traders had been using them to monitor client trading flows, build models, and produce written and graphic materials, and some are frustrated. Goldman hasn't said how many people lost access.
That's a turn from July, when CIO Marco Argenti praised "vibe coding" by non-technical staff. The October guidance followed a trial period that weighed efficiency gains against risk and controls. Goldman is still investing in AI. It's rolling out its in-house GS AI Assistant to about 46,000 employees and running agentic AI trials in trade surveillance.
Goldman isn't the only bank drawing lines:
Bank | Where it drew the line |
|---|---|
Goldman Sachs | Coding tools limited to engineers and strats; in-house assistant for everyone else |
JPMorgan Chase | Jamie Dimon warned of "rogue agents" and said outside AI agents won't be allowed inside the bank |
Bank of America | Limited agents' access to consumer deposits while it develops liability rules |
Citigroup | Reportedly doesn't let third-party agents log into accounts or make transactions |
Does your store have an AI policy your team actually knows about?
We see this as a signal to every business that handles sensitive data, and dealerships handle customer PII all day. As AI-driven hacks and data breaches grow, balancing speed with security will dominate more of the conversation. That also puts pressure on companies to build secure internal tools fast, so people aren't left with nothing once the outside tools get locked down.
"If you're not thinking about those governance policies and communicating with your staff [about] what's expected and what's an option, you should be doing that right now."
Kyle Mountsier
Goldman gave its non-engineers a sanctioned alternative, and that's the part worth copying. Decide which tools are approved, for which roles, and what data never goes into them. Then tell your team before a cutoff surprises them. Kyle and Daniel Kim will dig into dealership AI governance on the first episode of the upcoming AI for Dealers podcast, and the full conversation from SEMA is on today's ASOTU episode.
