Wall Street quietly kicks the tires on Anthropic’s Mythos as Washington pushes banks to test AI

It has been reported that Goldman Sachs, Citigroup and several other big banks are quietly testing Anthropic’s Mythos model inside their walls, part of a broader push by financial firms to vet generative AI before it hits trading floors and client desks. It has also been reported that JPMorgan Chase is the only bank named in Project Glasswing, a government-linked effort that — according to reporting — aims to coordinate AI testing with financial institutions. Slow and careful. Curious. A little unnerving.

What banks are doing — and what Project Glasswing is

Details remain thin. Bloomberg’s reporting says internal trials are underway at multiple firms as they probe Mythos for research, automation and risk tools, it has been reported that U.S. agencies are encouraging banks to run controlled tests first. Project Glasswing, allegedly a government initiative to evaluate AI’s safe use in financial systems, apparently lists JPMorgan as the only publicly named participant so far. That’s the kind of name-check that raises eyebrows — for reputational reasons if nothing else.

Why this matters now

There’s a tug-of-war in plain sight: firms want the efficiency gains of next‑generation models, regulators want assurance that those models won’t hallucinate trades, leak data, or amplify systemic risk. Who moves first? Who gets burned? Banks don’t like being guinea pigs, but neither do depositors who expect their money to be handled with a human-level safety net. It’s Silicon Valley rollouts colliding with Wall Street caution — a cultural mashup if ever there was one.

Expect more pilots, more guarded disclosure, and more pressure from Washington for standardized testing frameworks. The real story will be what comes next: clear rules, usable guardrails, or messy headlines. Either way, the pressure to prove these models can be trusted is no longer theoretical — it’s on the clock.

Sources: bloomberg.com