Anthropic Launches 10 AI Agents to Automate the Financial Sector
Anthropic has introduced 10 specialized AI agents based on Claude for banks and insurance companies, capable of automating routine tasks including auditing, report preparation, and pitchbook creation.
Anthropic didn't just release a set of tools — it is methodically cracking open the financial sector, which has relied for decades on opacity and manual labor. While everyone debates whether AI will replace traders, the real revolution is happening in the back office, and its consequences will be far more disruptive to Wall Street's established hierarchy.
The Core: What's Really Happening
Anthropic's announcement of 10 agents for finance is not a software release but a strategic manifesto. The company is moving toward becoming not just an API provider for banks, but their operating system. The key signal is the direct integration of Claude into Excel, PowerPoint, and Word with context preservation across applications. This is not a chatbot that needs to be told what EBITDA is. This is an agent that builds a financial model in Excel from uploaded reports, then packages the findings into a pitchbook in PowerPoint — all without breaking the thread of meaning. Technically, this means Claude Opus 4.7 has learned not just to generate text from a prompt, but to execute multi-step sequences of actions while switching context between different document types. This is the shift from "generative AI" to "executive AI."
The simultaneous announcement of the agents along with the creation of a service company with Blackstone, Hellman & Friedman, and Goldman Sachs with a budget of around $1.5 billion are two parts of one plan. The first part is the product (10 templates). The second is a deployment channel for mid-sized companies that lack the resources to hire McKinsey. The figure of 300,000 business clients and annual revenue that jumped from $9 billion to $30 billion in a few months indicates that the financial sector is already voting with its wallet, and this launch merely formalizes what banks were already doing in a makeshift way.
Timeline and Context
The story of this launch didn't start in May 2026. In July 2025, Anthropic first launched Claude for Financial Services with limited access to FactSet. By October, Claude was already working in Excel, able to create coverage reports. In February 2026, the Opus 4.6 model sharply improved financial reasoning and spreadsheet handling. And now, on May 5-6, 2026, there are 10 pre-configured agents capable of performing specific roles: Pitch builder, Earnings reviewer, General ledger reconciler, KYC screener.
In parallel — and this is critically important — on May 4, the creation of a service company with the largest private equity players led by Blackstone was announced. And in the same week, news of a partnership with FIS to create agents for detecting financial crimes. These are not isolated events. This is a coordinated blitzkrieg on three fronts: ready-made tools for the mass market, a service structure for deployment in mid-sized businesses, and strategic alliances with infrastructure players like FIS.
Who Wins and Who Loses
Winners:
- JPMorgan Chase and the largest banks. Jamie Dimon's presence at the presentation is not just a courtesy. Banks of JPMorgan's caliber, with $4.4 trillion in assets, get the opportunity to cut thousands of person-hours on pitchbook preparation, report auditing, and KYC checks. Their gain is not in headcount reduction but in speed: the pitch builder does in hours what a team of analysts did in weeks.
- Private equity funds. Blackstone, Apollo, Carlyle are already deploying Claude for deal analysis. For them, this means the ability to evaluate acquisition targets faster than competitors.
- Anthropic itself. The launch solidifies its position ahead of an IPO that could take place as early as October 2026. Revenue growth of 80x year-over-year and a valuation above $900 billion in the latest round make the company the primary beneficiary.
Losers:
- FactSet and traditional data providers. FactSet shares fell 8.5% on the day of the announcement. Anthropic's agents connect directly to dozens of data sources via connectors — Dun & Bradstreet, Moody's, S&P Capital IQ. This means value is shifting from the data itself to the agent that interprets it. FactSet risks becoming just one of the pipes connected to the Claude system.
- Legacy financial software vendors. Anthropic CEO Dario Amodei did not mince words: some SaaS companies will "lose market value, go bankrupt, be completely ruined." This is not a forecast but a warning to those who do not integrate into the Claude ecosystem.
- Junior financial analysts. "Junior analysts cooked" — the phrasing is harsh but accurate. Tasks that once required armies of MBA graduates: gathering comparables, reviewing earnings reports, reconciling models — are now automated. This does not mean immediate layoffs, but the career trajectory for entry-level finance professionals is narrowing.
What the Media Isn't Saying
The main non-obvious insight concerns the security mechanism, which is actually a Trojan horse for monopolization. Anthropic has built a "human-in-the-loop" system into the agents: the user must confirm results, and everything is logged. This is presented as a compliance solution for regulators. But the real function of this mechanism is to create the largest database in history of how financial decisions are actually made.
Imagine: thousands of banks and funds use Claude for auditing, risk assessment, and pitchbook creation. Every time an analyst confirms or rejects an AI output, the system records that signal. After six months, Anthropic will have a labeled dataset showing what "correct" and "incorrect" financial judgments look like from the perspective of real professionals. This is data that neither Bloomberg, nor Reuters, nor any regulator has. This is intellectual rent that makes Claude invulnerable to competitors: every user transaction improves the model for free. Regulators don't see this yet because they view AI as a tool, not as a system for accumulating precedent knowledge.
The second underestimated factor is the role of FIS. The partnership with FIS, whose systems process payments for thousands of banks, gives Claude access not just to financial reports but to real transaction flows. This transforms the AML agent from a simple list checker into a system that sees the entire landscape of money movement and can find patterns invisible to isolated banking systems.
Forecast: Next 30 Days and 90 Days
30 days (by early June 2026):
A wave of announcements from competitors will follow. OpenAI will try to hastily announce similar financial agents, citing GPT-5.5, but the gap in benchmarks is already noticeable: Claude Opus 4.7 scored 64.37% on the Vals AI Finance Agent benchmark versus 59.96% for GPT-5.5. Microsoft, which owns a stake in OpenAI, will find itself in an ambiguous position because Claude's integration with Excel and PowerPoint makes Microsoft's product more valuable, even if it's a competitor's product. I expect shares of FactSet and other niche financial data providers to continue declining another 5-10%. Large banks will start closed pilots with the agents, but there will be few public case studies — no one will want to reveal the scale of automation before quarterly earnings.
90 days (by August 2026):
The Anthropic-Blackstone service company will begin its first commercial projects with mid-sized banks and insurance companies. This will trigger a chain reaction among regional financial institutions that see the risk of falling behind. The key moment: the FIS partnership will start to bear fruit, and we will hear about the first major case where Claude's AML agent uncovered a money laundering scheme missed by traditional systems. By the end of the quarter, I expect at least three of the ten largest US investment banks to announce significant reductions in hiring of entry-level analysts. And most importantly: some regulator — likely the SEC or FINRA — will for the first time issue an official inquiry into how exactly Anthropic's agents work, what data they accumulate, and who is responsible for an AI error in financial reporting. This will be the first phase of a regulatory tech showdown that will define the boundaries of AI in finance for a decade to come.
— Editorial Team
No comments yet.