Wall Street is hoarding humans to audit AI
Unlike some sectors that call AI a reason to cut headcount, parts of Wall Street are deliberately keeping people on staff to audit and oversee AI systems as a compliance control. (x.com) That approach raises recruiting priorities toward candidates who can work with and check AI, not just be replaced by it. (x.com)
Wall Street spent 2024 and 2025 telling employees to use artificial intelligence faster, then turned around and kept real people on payroll to check what the machines were doing. JPMorgan Chase said its internal large language model platform reached 200,000 onboarded users within eight months, while Goldman Sachs rolled out its own assistant firmwide after about 10,000 employees had already tested it. (jpmorganchase.com) (usnews.com) That sounds backwards until you remember what a bank actually sells. A bank can survive a slow spreadsheet, but it cannot survive a regulator deciding its models were unchecked, undocumented, or unfair. (federalreserve.gov) (occ.treas.gov) The rulebook Wall Street keeps reaching for is older than ChatGPT. Federal Reserve guidance from April 4, 2011 says model risk management needs strong validation, governance, policies, and controls, which is regulator language for “someone has to inspect the machine before the machine can inspect anything else.” (federalreserve.gov 1) (federalreserve.gov 2) The Office of the Comptroller of the Currency said in 2022 that bad outcomes can come from faulty data, weak testing, or limited human oversight. That means an artificial intelligence tool in lending, surveillance, or compliance creates a second job next to the first job: the human who has to challenge it. (occ.treas.gov) The National Institute of Standards and Technology built the same idea into its Artificial Intelligence Risk Management Framework. Its four functions are govern, map, measure, and manage, which is less “replace the analyst” and more “hire people who can watch the autopilot and know when to grab the wheel.” (nist.gov) (airc.nist.gov) That is why some Wall Street hiring is splitting in two directions at once. One lane is fewer classic entry-level jobs as software drafts slides, summarizes calls, and searches documents faster; the other lane is more demand for people who can validate outputs, document workflows, test bias, and explain decisions to compliance teams and examiners. (cnbc.com) (federalreserve.gov) Morgan Stanley’s wealth unit gives a clean example of the split. Its artificial intelligence tools were built to retrieve firm knowledge for advisors and draft meeting follow-ups, but those tools sit inside a business where every client note, recommendation, and record can become a supervision issue if the output is wrong. (businesswire.com) (cnbc.com) So the new “safe” Wall Street candidate is not just the person who can use artificial intelligence prompts. It is the person who can show where the data came from, describe why the answer might be wrong, and leave an audit trail a regulator can read six months later. (federalreserve.gov) (nist.gov) In other words, parts of finance are not hoarding humans because the machines failed. They are hoarding humans because once the machines are inside the bank, the humans become the control system that lets the bank keep using them at all. (occ.treas.gov) (moodys.com)
