JPMorgan adds Claude spending limits and a new security setup for some engineers

JPMorgan has launched $2,000 monthly spending limits for some employees using Claude Code, according to internal Teams messages seen by Business Insider. The messages referenced errors including “ExceededBudget” and “Budget=2000.0,” with employee replies indicating that costs reset monthly and that some users can request higher limits.

The move is notable because Business Insider reports it is the first indication of AI spending or token caps at the bank, which has previously tracked engineers’ AI usage.

The cap comes as token spending becomes a more visible cost pressure across Wall Street. JPMorgan’s nearly $20 billion annual tech budget gives it room to experiment, but the article frames the new limits as part of a more disciplined approach to tying AI costs to measurable business value.

A JPMorgan spokesperson declined to comment on specific token limits or the bank’s cost strategy, but said the firm considers metrics including adoption, productivity improvements, speed, risk reduction, and business impact.

The spending limits coincide with a broader change in how some JPMorgan engineers use Claude through a new coding architecture called Devspace. Business Insider reports that Devspace is a containerized, sandbox-like server hosted in Amazon Web Services, designed to restrict Claude’s access to employee credentials and internal systems.

The setup is especially relevant for AI agents, which can make decisions and carry out tasks independently. JPMorgan’s security approach aims to isolate AI activity from sensitive desktop environments that may contain emails, files, and personal information.

Business Insider reports that only a sliver of JPMorgan’s 65,000-person Global Technology division currently has access to Claude. An internal dashboard showed around 8,000 people had Claude licenses in August, while a Devspace-related Teams group had about 1,900 members in early September.

The rollout highlights a core enterprise AI challenge: scaling coding assistants while controlling costs, credentials, and risk. For banks, where customer data and trust are central, security architecture may become as important as model access itself.