Latest / MBA Training AI / Goldman Sachs AI policy: build the tool first

Transcript

Goldman Sachs AI policy: build the tool first · MBA Training AI

• 0:00
  Host: Welcome to the Leaders' Insights podcast.
• 0:07
  Host: Today's episode, how Goldman Sachs built an AI usage policy that employees actually
• 0:12
  Host: followed.
• 0:13
  Host: Most companies have an AI policy nobody reads.
• 0:16
  Host: Goldman Sachs claims theirs actually changed how people work.
• 0:20
  Host: Why should I believe that isn't just a press release?
• 0:23
  Expert: Because the evidence is in the usage numbers, not the pros.
• 0:27
  Expert: By early 2026, Goldman had their internal assistant, they call it GSAI Assistant, in
• 0:33
  Expert: the hands of roughly 46,000 employees.
• 0:36
  Expert: When adoption is that broad, the policy isn't decoration, it's load-bearing.
• 0:40
  Host: Fine, but adoption isn't compliance.
• 0:43
  Host: Plenty of people use tools while quietly breaking every rule attached to them.
• 0:48
  Expert: Right, and that's exactly what most firms get wrong.
• 0:51
  Expert: They write a policy as a list of prohibitions, don't paste client data, don't use
• 0:55
  Expert: unapproved tools, and then act shocked when it's ignored.
• 1:00
  Expert: A ban you can't detect is a suggestion.
• 1:03
  Expert: Goldman flipped it.
• 1:04
  Expert: Instead of telling people what not to do outside the walls, they made the sanctioned
• 1:08
  Expert: path the easiest path.
• 1:11
  Host: Meaning what concretely?
• 1:12
  Expert: Meaning they built the guardrails into the tool itself.
• 1:15
  Expert: The assistant runs inside their own environment, so sensitive data never leaves.
• 1:20
  Expert: If the only good AI on your desk is the compliant one, you don't need a 30-page
• 1:25
  Expert: document policing shadow AI.
• 1:27
  Expert: The ungoverned tools employees smuggle in on their own.
• 1:31
  Host: That sounds convenient for Goldman.
• 1:33
  Host: They have a few billion dollars to build in-house.
• 1:36
  Host: What's a normal company supposed to do with that lesson?
• 1:38
  Expert: The transferable part isn't the spending, it's the sequence.
• 1:43
  Expert: Most teams write the policy first and build capability never.
• 1:47
  Expert: Goldman gave people a real tool, then wrote rules that matched how the tool actually
• 1:51
  Expert: worked.
• 1:52
  Expert: The rule was legible because it described reality.
• 1:55
  Expert: That costs judgment, not money.
• 1:58
  Host: Let me push on the employees actually followed claim.
• 2:01
  Host: Financial firms are terrified of regulators.
• 2:03
  Host: Isn't fear doing the real work here, not clever policy design?
• 2:08
  Expert: Fear gets you performative compliance, people who technically obey, and quietly resent
• 2:13
  Expert: it.
• 2:14
  Expert: That's how you get a workforce that stops volunteering the risky but valuable use
• 2:18
  Expert: cases.
• 2:19
  Expert: What Goldman did differently was tiered permission.
• 2:23
  Expert: Not every task carries the same risk, so they didn't govern them identically.
• 2:28
  Expert: Summarizing a public earnings call is not the same as drafting client-facing advice,
• 2:34
  Expert: and treating them the same is how you train people to ignore you.
• 2:37
  Host: Give me the mechanism.
• 2:38
  Host: How does a bank decide which task sits in which tier?
• 2:42
  Expert: They classify by data sensitivity and by whether output goes to a human first.
• 2:47
  Expert: Internal research summary, low stakes, wide access, anything touching client money or
• 2:53
  Expert: regulated advice, tighter controls plus a human review step, a person signs off before it
• 2:59
  Expert: goes out.
• 3:00
  Expert: The genius is mundane.
• 3:02
  Expert: Employees can see why the strict cases are strict, so the strictness feels earned rather
• 3:06
  Expert: than arbitrary.
• 3:07
  Host: You keep praising Goldman.
• 3:10
  Host: What did they get wrong?
• 3:11
  Expert: They were slow, deliberately, and that annoyed people internally.
• 3:16
  Expert: They ran controlled pilots with their engineers and select teams before the wide rollout, which
• 3:21
  Expert: meant plenty of staff watched competitors move faster and felt left behind.
• 3:26
  Host: I'd argue that's a feature.
• 3:28
  Host: A policy nobody's tested is a rumor with a logo, but it's a real cost, and anyone
• 3:33
  Host: copying this should expect the impatience.
• 3:36
  Host: Isn't there a risk the whole thing becomes a comfortable cage?
• 3:40
  Host: People only think inside the sanctioned tool and stop exploring?
• 3:43
  Expert: That's the honest danger, yes.
• 3:46
  Expert: This can calcify into a place where good ideas go to fill out a form.
• 3:50
  Expert: The counter is to keep a fast lane for proposing new use cases, a request path that gets answered
• 3:55
  Expert: in days, not a committee that meets quarterly and adjourns before deciding anything.
• 4:00
  Host: So if I run a midsize team and I want one thing I can do Monday, what is it?
• 4:05
  Expert: Stop writing the policy first.
• 4:07
  Expert: Pick one approved tool.
• 4:08
  Expert: Put it in front of people.
• 4:10
  Expert: Watch how they actually use it for two weeks.
• 4:12
  Expert: Then write rules that describe what you saw.
• 4:15
  Expert: People only think that a policy should document reality, not fantasize about it.
• 4:19
  Expert: A rule people recognize is a rule people keep.
• 4:22
  Expert: Describe the terrain before you draw the map that I can use.
• 4:25
  Host: That's your five.
• 4:26
  Host: The full AI track, structured no fluff, is at MBA-training.com.