AI Finance track opens at AIE NYC; OpenAI, Anthropic invest in banking plugins

Latent Space released a full Finance track today for the second annual AIE NYC conference, scheduled for October, with Early Bird Tickets opened and speaker applications remaining open. The track features ten speakers from major financial institutions and fintech firms addressing the practical deployment of AI agents in financial services.

FactSet's Yogendra Miraje framed the core challenge: at a company serving thousands of financial-data clients, "AI skills" are not mere features but require ownership, search, evals, audits, and governance to become enterprise-grade agent infrastructure. Nubank, a digital bank with 100M+ customers, has discovered that simulations can turn agent evals from a bottleneck into a release mechanism for shipping customer-facing AI faster. Intuit, serving ~100M consumers, small businesses, and accountants, noted that generic LLMs are insufficient; finance AI must understand real state, actions, outcomes, and risk. Kepler, which indexes millions of filings and market documents in financial research, emphasized that "verifiable AI" means every answer needs provenance, reconciliation, and review. Nubank's Lucas Palma raised supply-chain security: vetting thousands of AI skills before developers use them becomes a supply-chain security problem, not just a developer experience problem. Morgan Stanley's Brendan Hogan Rappazzo stated that inside a global financial institution managing trillions in client assets, multi-agent research only matters if humans can trust the experimental environment it optimizes in. FlyersSoft's Divakar Kumar noted that event-sourced systems already preserve the historical trail financial agents need, making them a natural foundation for auditable production decision loops. Fidelity Investments' Sai Krishna Rallabandi pointed out that group-chat and wearable agents force new thinking around memory, permissions, and prompt-injection defense at an asset manager with trillions under administration. China Resources Holdings' Shawn Chan, speaking for a Fortune Global 500-scale conglomerate, emphasized that finance AI has to be built for the investment memo—reconciled numbers, uncertainty labels, and provenance beat demo polish. Auditoria AI's Ramana Siddanth Emani framed back-office finance automation: the bottleneck may be the developer loop itself, with agents increasingly generating workflows while humans verify the financial truth.

In parallel, OpenAI and Anthropic released dedicated finance offerings. OpenAI held an NYC event with dedicated equity investing and investment banking plugins in Codex, signaling that banking workflows are now embedded in the company's agent platform. Anthropic's Financial Services team also held an NYC event and released Cowork and Claude Code agent templates covering every workflow in corporate finance. OpenAI also released Codex Security CLI, an open-source repository scanner for repos and CI/CD that can scan codebases, track findings across runs, verify fixes, and integrate security checks into pipelines; it is available via npm. Additionally, OpenAI announced a program giving 10,000 researchers initially, expanding to 100,000 by 2027, free access to frontier models including the GPT-5.6 family, with business-grade privacy and security and up to four collaborators per workspace. Latent Space made clear that the Finance track is the mainstage theme for the second annual AIE NYC, with speaker applications open and a very high bar for finance-focused submissions given the expected attendee list. For the West Coast, the outlet expects to announce AIE CODE soon.

The rise of AI in Finance marks a shift from experimental pilots to production infrastructure at scale. The newsworthy angle is not that AI is being used—it is that every major financial subsector now needs AI governance, not just features. FactSet, serving thousands of financial-data clients, frames this plainly: "AI skills" require ownership, search, evals, audits, and governance to become enterprise-grade agent infrastructure. Nubank, with 100M+ customers, has discovered that agent evals themselves can become the release mechanism for shipping customer-facing AI faster, turning a bottleneck into a pipeline. This pattern repeats across the Finance track: Intuit (serving ~100M consumers, small businesses, and accountants) finds generic LLMs insufficient; Kepler (indexing millions of filings) requires verifiable AI with provenance; Morgan Stanley (managing trillions in client assets) demands trust in the experimental environment multi-agent research optimizes in.

OpenAI and Anthropic's parallel moves—dedicated NYC events, banking plugins, agent templates—signal that both labs see Finance as a major subsector and are moving to embed themselves in banking workflows. OpenAI's decision to hand Codex Security CLI to the open-source community and to fund 100,000 academic researchers with free GPT-5.6 access frames the play as infrastructure-level adoption: security scanning is embedded in dev pipelines; research acceleration happens outside labs. Anthropic's Financial Services team and Claude Code templates take a similar infrastructure approach. The Finance track at AIE NYC, released today and featuring global institutions (Fidelity, Morgan Stanley, China Resources Holdings), crystallizes this: the conference is centering on a concrete vertical where AI agents are becoming mission-critical, and the governance / evaluation / audit / harness questions are no longer optional.