Claude for Financial Advisors Is Here. But Is the Industry Building AI Backward?

Written by: Kyle Wiggs

Anthropic just launched Claude for Financial Advisors.

The partner list: Schwab. Addepar. Wealth.com. BlackRock. Vanguard. Orion. SS&C Black Diamond. Envestnet. iCapital. Wealthbox. Zocks. Microsoft 365. Salesforce. DocuSign. Box. FactSet. S&P Global. Morningstar.

That's 18+ separate vendors being connected through one AI layer using something called MCP — Model Context Protocol. The distribution is massive. Claude inside Schwab Advisor Center reaching 16,000+ RIAs overnight is the single biggest distribution event advisor AI has seen. Wealth.com's estate document analysis with page-level citations is genuinely impressive work. I'm not here to pretend this isn't a real moment for the industry. It is.

But I want to say something nobody at Future Proof seems willing to say.

The entire industry is going about this backwards. And the risks are massive.

The premise — and the problem

The premise behind Claude for Financial Advisors is that you take a dozen standalone SaaS products — each built by a different company, each storing data in a different database, each with a different security model — and you stitch them together with AI after the fact. MCP is the protocol that makes the stitching possible. Pull CRM data from Wealthbox. Portfolio analytics from Addepar. Financial planning from Envestnet. Estate documents from Wealth.com. Model portfolios from BlackRock. Meeting notes from Zocks.

Every one of those calls crosses a vendor boundary. Every boundary is a compliance event in a regulated industry.

Who's actually holding the bag

Here's what every advisor needs to understand before they turn this on: read the fine print.

Every one of these vendor agreements — Anthropic, Schwab, Addepar, Wealth.com, all of them — will contain indemnification clauses that protect the vendor. When Claude pulls client PII from your CRM, cross-references it with estate documents from Wealth.com, and routes it through portfolio data from Addepar, the data is crossing three separate vendor boundaries in a single AI reasoning chain.

When something goes wrong — and in a system with 18+ integration points, something will go wrong — who's liable?

Not Anthropic. Not Schwab. Not Addepar. Not Wealth.com.

You. The RIA. The registered fiduciary.

You are the regulated entity. You hold the fiduciary duty. You are responsible for the due diligence on every vendor whose system your client data touches. And you just gave an AI permission to dynamically decide which of 18+ vendor systems to shuttle that data through at runtime. Newsflash: AI constantly makes mistakes and then holds you responsible.

Vendor contracts disclaim responsibility for breaches. They disclaim responsibility for AI-generated outputs. They disclaim responsibility for how their tools interact with other vendors' tools. Every single one of those disclaimers points liability in one direction: back at the advisor.

This isn't speculation. FA Mag's own reporting on this launch quotes compliance experts warning that firms without records connecting an AI system's inputs and outputs to the advisor who acted on them face supervisory gaps regulators will immediately identify. Unreviewed AI-drafted client communications can breach fiduciary disclosure duties if they turn out to be inaccurate. Entering client data into public AI systems can violate safeguarding requirements.

In April, Anthropic accidentally exposed the internal instructions behind its Claude Code product. They called it a release packaging issue caused by human error. But advisors rightly pointed out: they are responsible for sanitizing personal client information before it reaches any AI system. They should check vendor agreements for zero-data-retention terms. And when those agreements disclaim responsibility for breaches, it's the RIA holding the compliance bag.

With 18+ vendor boundaries, you're holding 18+ bags.

The read-only ceiling

Here's what else nobody in the hype cycle is saying out loud: every single capability described in these announcements — meeting prep, plan updates, analytics, follow-up drafts, document analysis — is a read operation.

Not one word about executing a trade. Generating an invoice. Rebalancing a portfolio. Running a tax-loss harvest.

Because Claude doesn't own those rails. It reads from systems it doesn't control. It can draft a meeting summary. It cannot do the work that meeting summary describes. The advisor still has to go execute across five different tabs.

That's the ceiling when you stitch standalone SaaS together with an AI layer. You get a smarter reading layer on top of the same fragmented stack — and you accept all the liability of 18 vendor integrations for the privilege.

A different architecture

At UX Wealth Partners, we made a different bet. One that is nearly complete, in January.

We built the trading engine. The billing engine. The risk engine. The planning engine. The CRM. The investment engine. The execution infrastructure. All first-party. One canonical data model from the firm level down to the individual tax lot.

We didn't build a TAMP and then bolt AI onto it. We built the entire operating platform knowing the AI was coming.

Now we're putting it inside. A native AI layer that doesn't need a protocol to discover tools across 18 vendors — because there are no vendors. It sits on top of everything we built. Client data stays within a single architecture. The compliance trail isn't bolted on after the fact. It's the architecture itself.

One platform. One vendor agreement. One audit trail. One place where liability lives and can actually be managed.

And because we own the full execution path — our own OMS, our own FIX engine, our own billing engine — our AI isn't limited to reading your book and handing you a summary to go execute yourself across five different tabs.

The architecture is built so it can propose an action, the advisor authorizes it, and the platform executes. Trading. Billing. Rebalancing. On rails we built and control. Full audit trail, native. Not across 18 vendor boundaries. Across zero.

We're shipping this January 1st. Not talking about it at a conference. Building it now.

Access vs. architecture

I'm not taking shots at Schwab, Anthropic, Addepar, or Wealth.com. What they shipped is a genuine step forward for advisor access to AI. The distribution alone moves the industry.

But access and architecture are different problems. Solving one doesn't solve the other.

The rest of the industry is hoping AI can stitch their standalone SaaS together into something that feels like a platform. We built the platform first. That's not a feature difference. That's an architectural difference. And in regulated wealth management, architecture is the only thing that scales.

Connecting AI to a platform is a feature. Making AI the platform is a company.

MCP is a brilliant solution for connecting disconnected systems. We just don't have disconnected systems.