OpenAI has signed a strategic partnership with JPMorgan Chase that will put its Codex coding agent and its cybersecurity models to work inside the largest bank in the United States. OpenAI president Greg Brockman revealed the deal on November 4 at the Wall Street Journal’s technology conference in Napa, California, describing it as newly signed and not previously announced. Neither company has disclosed financial terms, a contract length, or a start date. The bank plans to use the tools across its engineering and security teams, and Brockman said the two companies will also explore wider uses for the bank’s employees and customers.
What the deal covers
The agreement covers two OpenAI products. The first is Codex, the company’s agentic coding tool, which writes and edits code inside a developer’s project. The second is a set of specialised cybersecurity models, delivered through Daybreak, the programme OpenAI uses to give approved organisations access to its advanced security capabilities, tools and support.
Brockman gave no further detail on which cyber models the bank will receive. The newest candidate is GPT-6 Cyber, the security model OpenAI released on November 5, which finds software flaws and builds working exploits for approved defenders. OpenAI has not confirmed that JPMorgan will get that model, and the bank has said nothing publicly about the deal at all.
Brockman confirmed the deal himself after the event, writing that the two companies will explore how advanced AI can support the bank’s engineers, security teams, employees and customers. He added that he was especially excited about working together on cybersecurity, to help secure and protect the financial system to a level that is only possible with AI.
A bank that already runs on AI
This is not the bank’s first encounter with OpenAI’s models. JPMorgan has offered them to staff since 2024 through LLM Suite, its internal platform, which also carries Anthropic’s models. Crypto Briefing reports about 150,000 employees used the platform each week by early 2026, and the bank has said its AI work generated roughly $2 billion in annual savings.
The bank has also kept a second lab on its books. Since April 2026 it has worked with Anthropic on Project Glasswing, a defensive cybersecurity effort that brings organisations together to protect software. That work continues alongside the new OpenAI agreement, so the two labs are now competing for the same bank’s security budget rather than one replacing the other.
Our Read
The deal matters less for what it says about JPMorgan than for what it says about OpenAI. A regulated bank running an agent that writes code and a model that finds exploits is the hardest possible reference customer, and Brockman chose to announce it on a conference stage rather than in a press release.
That choice is a sales pitch, aimed at every other bank and every security chief watching. OpenAI has spent the autumn arguing that its enterprise business now outweighs its consumer business, and a named customer of JPMorgan’s size is the evidence that argument needs. The question is whether the bank’s engineers adopt Codex in practice, which no announcement can prove.
The cybersecurity half of the deal is the more consequential part. Banks are the most heavily defended targets in the world, and letting an outside lab’s models probe their systems requires a level of trust that no vendor slide can manufacture.
If JPMorgan’s security teams get real value from OpenAI’s cyber models, other banks will follow quickly. If the models produce noise or false alarms, the deal will quietly shrink, and OpenAI will have spent its best reference customer on a pilot that never grew.
Why It Matters
Wall Street’s largest bank putting an AI coding agent and offensive-capable security models into daily use is a signal to the whole financial sector. Regulated industries have been the slowest to let AI agents touch production systems, and a JPMorgan deployment gives every rival bank a precedent to point at in its own boardroom.
It also sharpens the contest with Anthropic, which already works with the bank on defence and now has to share the account. For OpenAI, the deal is a named customer for an enterprise story it has been telling all autumn; for the bank, it is a bet that AI can guard the financial system better than people alone.
Background
OpenAI launched Codex as its agentic coding tool and has spent 2026 pushing it into enterprise engineering teams, including by acquiring Pierre Computer’s Code Storage system. Its Daybreak programme vets organisations for access to its advanced cybersecurity models. Brockman said at the same conference that OpenAI could be ready to go public in 2027.
Source: PYMNTS report · Crypto Briefing report
0 Comments