JPMorganChase has signed a contract to use OpenAI’s Codex agents to write, review, debug and ship code, and OpenAI models for cybersecurity. The deal gives OpenAI a foothold in two consequential areas of work at one of the world’s largest banks, though the contract’s terms and the models used beyond Codex remain undisclosed.
OpenAI Watch analysis
What happened
OpenAI president Greg Brockman announced the contract, according to American Banker’s report. A JPMorganChase spokeswoman told the publication that the bank uses a variety of tools and will continue to use multiple providers. She declined to disclose the contract terms or which OpenAI models the bank will use apart from Codex.
The bank already uses other AI tools, including Anthropic’s Claude for engineering and internal analysis, and GitHub Copilot among its developers, the report says. JPMorganChase chief financial officer Jeremy Barnum said last month that the bank spends “something like $700 million annually” on AI, while noting that benefits had not yet appeared in quarterly results.
Why it matters
This is a substantial enterprise deployment, not just another staff chatbot. Coding agents are being used for software development, while cybersecurity puts AI into a more sensitive class of work. The bank’s continued use of multiple providers also makes clear that this is not, at least from its public account, an exclusive bet on OpenAI.
JPMorganChase’s scale could make its choices influential for other banks considering how to deploy AI. But the contract itself does not show that the tools improve productivity or security. Barnum said some experiments rewriting minor legacy tools had not worked perfectly, and that any gains would have to be weighed against costs such as tokens.
Our read
The cybersecurity use is the more consequential part of this announcement: helping write code is one thing; involving a model provider in securing a bank is another. That makes the practical questions unusually important: what systems the models can access, how their work is checked, and how the bank measures results. The headline says deployment; the proof of value is still to come.
What to watch
- Whether JPMorganChase discloses which OpenAI models it uses and where they are deployed.
- How the bank evaluates the security, reliability and cost of AI-assisted work.
- Whether it reports measurable productivity gains, rather than activity alone.
Discussion spark: Should banks let AI providers help secure their systems, or is that a level of trust that ought to require unusually strong evidence first?
Sources and evidence
- Source update (10 November 2026, 18:51 UTC)
OpenAI Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by OpenAI.