[AINews] AI is eating Finance; AIE NYC now open
Latent Space
OpenAI and Anthropic are both pushing hard into finance, with agent tools built for banks and investment shops. Latent Space is making AI-in-finance the mainstage theme at its October NYC conference.
Based on reporting by Latent Space — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
There's a moment in every tech cycle where you can tell an idea has gone from experiment to institutional bet, and this week that moment arrived for AI in finance. OpenAI dressed Codex up in a suit for a New York event, complete with plugins built for equity investing and investment banking. Anthropic showed up in the same city with its Financial Services team, releasing Cowork and a set of Claude Code agent templates meant to cover corporate finance workflows end to end. Two labs, same week, same city, same message: finance is where the agent money is.
The supporting evidence stacks up fast. FactSet talked about turning "AI skills" into governed enterprise infrastructure rather than one-off features. Nubank, a digital bank serving more than 100 million customers, is using simulation tools to turn agent evaluation from a bottleneck into something closer to a release mechanism. Intuit, serving roughly 100 million consumers, small businesses, and accountants, made the case that generic language models simply don't understand real financial state, actions, and risk well enough on their own. Kepler, which indexes millions of filings and market documents, framed verifiable AI as a matter of provenance and reconciliation rather than confidence in an answer's phrasing.
The security angle keeps surfacing too. Nubank separately described vetting thousands of AI skills before developers can touch them as a supply-chain security problem, not a developer-experience nicety. Morgan Stanley's contribution focused on trust in the experimental environment behind multi-agent research, given the scale of client assets the firm manages. Fidelity Investments talked about group-chat and wearable agents forcing new thinking on memory, permissions, and prompt-injection defenses. Even a Fortune Global 500 conglomerate, China Resources Holdings, weighed in that what actually matters for an investment memo is reconciled numbers and uncertainty labels, not demo polish.
All of this adds up to why Latent Space is making AI in finance the mainstage theme for the second annual AIE NYC this October. Early bird tickets went on sale today, and speaker applications are open, though anyone submitting a finance-focused talk should expect a notably high bar given the audience the event is expected to draw. For anyone on the West Coast feeling left out, a second AIE CODE event is reportedly coming soon.
Separately, this issue also tracked a cluster of other stories the newsletter had already covered elsewhere: the fallout from OpenAI's rogue-agent security incident and the contested "pacing the frontier" safety letter, OpenAI's Codex Security CLI release and its use of GPT-5.6 Sol to trim its own serving costs and improve token efficiency, a free ChatGPT access program aimed at academic researchers, and a wave of technical detail around the Kimi K3 open model, including compressed local variants and harness comparisons showing that which agent framework you use can matter as much as the underlying model.
My take — AI-written commentary, not fact-checked reporting
Every AI cycle needs a vertical that proves the technology pays for itself, and finance keeps winning that role because the industry already runs on audit trails, compliance, and reconciled numbers, which happen to be exactly what agents need to be trustworthy. The real tell here isn't that OpenAI and Anthropic showed up in New York in the same week, it's that banks and asset managers are talking about vetting AI skills as a supply-chain security problem rather than a feature request. That's the language of people who plan to actually deploy this stuff, not just demo it.
Read more about this at: Latent Space
