Chronicle, May 19: Anthropic's enterprise week, in three movements
Wall Street briefing on the 5th. Code w/ Claude SF on the 6th. KPMG global alliance on the 19th. The three pieces fit together more tightly than the headlines suggest.
Three Anthropic announcements in two weeks. Read alone, each is an enterprise milestone. Read together, they are the same story.
May 5 — Wall Street, New York. Anthropic disclosed Claude in production at JPMorgan Chase, Goldman Sachs, Citi, AIG, and Visa. It debuted Claude Opus 4.7, which leads Vals AI’s Finance Agent benchmark at 64.4%. Jamie Dimon, on the record. Peter Zafino, on the record. Lori Beer, on the record. The accuracy claim was 88% on insurance claims out of the box at AIG.
May 6 — Code w/ Claude, San Francisco. Managed Agents shipped. Claude Code and Opus API rate limits raised. Self-hosted sandboxes in public beta. MCP tunnels in research preview. 20+ legal MCP connectors. 12 legal practice-area plugins. Project Glasswing’s Claude Security beta opened. Then it traveled — London on May 20–21, Tokyo on June 5–6.
May 19 — KPMG, Amstelveen. Global alliance. 138 countries. 276,000 employees. Claude embedded in Digital Gateway, tax and legal tooling, cybersecurity, IT modernization, and the firm’s private-equity portfolio work. Bill Thomas on the strategy. Daniela Amodei on the scale: “that’s what a firm-wide commitment to AI looks like.” PwC had already expanded its own alliance earlier in the month.
The throughline: Anthropic is not pitching Claude as a model anymore. It is pitching Claude as the substrate of a regulated-industry professional-services stack, and it is naming the operators who have already bought in. The model is the wedge; the deployment count is the product.
What to watch next: whether competing labs disclose comparable production rollouts in the next two months. The gap between “in production” and “in disclosure” is its own kind of moat.