Wall Street Just Locked Out AI Agents. Is Your Enterprise Next?

Today, we're diving into:

Wall Street Just Locked Out AI Agents. Is Your Enterprise Next?

For a year, the industry has sold the same promise: your personal AI agent will book, buy and pay on your behalf. Last week, two of the biggest names in American finance said not with our customers' money.

Charles Schwab CEO Rick Wurster told Bloomberg Television that the firm is blocking unauthorized outside AI agents, including Meta's Muse, from moving client money. JPMorgan's Jamie Dimon had already said on the bank's earnings call that outside agents will not act inside its systems.

Is Your Data Ready For Agents That Move Money?

Find out whether your access rules, lineage and ownership can tell an agent what it may read and what it may never touch.

Blocked In The Tens And Twenties Every Week

This is not a hypothetical policy. Wurster receives a weekly report on blocked attempts, and he says large agents are stopped "in the tens and twenties" each week.

The worry is simple. Clients hand their password to an agent, and the agent then logs in and acts as if it were them.

That last question has no clear answer in the US yet. According to Forkast, no federal statute governs agent liability in financial services, and several agent liability bills have stalled in committee.

So The Banks Are Writing The Rules Themselves

Without a statute, firms are setting liability through exclusion. Forkast calls it "private ordering": if you cannot assign blame, you do not let the agent in.

Read Access Yes. Money Movement No.

The more interesting move is what Schwab is building instead. It plans to run Model Context Protocol (MCP) servers that let a client's chosen AI connect to Schwab's systems under Schwab's rules.

JPMorgan follows a similar line. Some third-party tools can already view account data, but they cannot send money or pay bills.

This Fundamentally Changes Your Agent Strategy

Most leaders will read this as a banking story. That is the wrong reaction. Every enterprise with customers, suppliers or partners will soon face agents at its own front door, and the same question will land on your desk.

The real bottleneck is not the agent. It is whether your systems can tell the difference between permission to read and permission to act.

What You Should Do Before Agents Knock On Your Door

Building that kind of control starts with trusted enterprise context. DataManagement.AI helps organizations create that foundation by unifying metadata, lineage, governance, business glossaries and knowledge assets into a single intelligence layer, so you know which data is sensitive, who owns it and which agents should ever be allowed near it.

Microsoft Reportedly Just Walked Away From Nscale's UK AI Data Center

Every AI roadmap quietly assumes the compute will be there when you need it. A story out of Essex this weekend is a reminder that the compute depends on something far less glamorous: the power grid.

Microsoft has reportedly withdrawn from a deal to anchor Nscale's first British data center in Loughton, Essex, according to The Telegraph, as cited by Bloomberg. Nscale has been told not to expect a power supply at the site until the 2030s.

Weeks Before An IPO

The timing is awkward. Nscale is seeking to raise up to $3 billion in an IPO, and its filing shows Microsoft and Anthropic account for about 85% of its contracts.

Nscale says it remains committed to Essex and is looking at other power sources and UK locations. Microsoft did not comment.

Not An Isolated Wobble

Nscale's Norway flagship is running about six months behind schedule. CoreWeave shares fell 25% the week before, and Australia's Firmus scrapped its IPO after failing to win enough support. Investors are getting wary of "neoclouds" that borrow heavily to build AI capacity.

Why This Lands On Your Desk

You may never sign a contract with Nscale. But the AI services you buy run somewhere, and more of that somewhere is being built by a handful of fast-growing, heavily financed providers.

This is not a vendor problem. It is a concentration problem. When a few buyers, a few builders and a stretched grid all have to line up, a delay in one place can ripple into the capacity, pricing or region your workloads depend on.

The better alternative is to treat compute like any other critical supply chain: diversified, documented and stress-tested.

DeepSeek Says Its New Open Model Finishes Agent Tasks For 20 Cents

While the West argues about power and permissions, the price of getting work done by AI keeps falling. The latest push came from China at the end of last week.

DeepSeek released V4.1-Flash-1017, an update aimed at faster agent and coding work with shorter outputs. The weights are open on Hugging Face under the MIT license, with a context window of 1 million tokens.

The Number That Matters Is Per Task

DeepSeek's headline is not a benchmark score. It is cost per completed task. On the DeepSWE coding benchmark, the company says the model scored 77.0% at an average of just under $0.20 per task.

DeepSeek is also opening its search engine to developers through Search and Fetch APIs, priced at $0.75 per 1,000 calls for each endpoint.

Read The Fine Print Before You Switch

Benchmarks capture only part of real-world quality, and developers often report a gap between scores and the code they actually get. The Decoder also reports that China plans to tighten rules on publishing open-weight models, which could affect future releases.

For regulated enterprises, data residency, model provenance and security review still decide whether a model is usable at all, however cheap it is.

The Wrong Reaction Is To Chase The Cheapest Model

A price gap that wide tempts teams to rip and replace. The better alternative is to know, workload by workload, what a completed task actually costs you today and what quality you need.

Know Your Inference (KYI) is becoming a core enterprise capability. By understanding how every inference affects cost, latency, accuracy and governance, you can see what each model really costs per completed task, route each workload to the right model, and negotiate with evidence instead of instinct.

Your opinion matters!

Hope you loved reading our piece of newsletter as much as we had fun writing it.

Share your experience and feedback with us below 'cause we take your critique very critically.

Thank you for reading

-Shen & Towards AGI team