BankingNewsAI Daily Brief  · 

TD deployed agentic AI for mortgage pre-underwriting, cutting pre-adjudication to under three hours.

🏦 3 Banking AI🤖 2 General AI

Banking AI

Financial institutions & fintech technology

3 stories
pymnts.com

TD put an agentic AI model into mortgage/HELOC pre-underwriting and cut pre-adjudication from ~15 hours to <3

TD says it has launched its first agentic AI model to pre-adjudicate real-estate secured lending files, shrinking the time before an underwriter can start from roughly 15 hours to under three. This is not a pilot claim; TD is positioning it as an enterprise milestone for end-to-end lending workflow automation.

Action

Benchmark your own mortgage/HELOC intake SLA against TD’s new baseline and fund the bottleneck-killers (document intake, data extraction, policy/rule application, exception routing). Put model risk, fair lending, and audit-trail requirements into the workflow now—agentic speed without defensible decisions will not survive second-line review.

Read article →
au.marketscreener.com

JPMorgan says AI tools are now rolled out globally across investment banking workflows

A senior JPMorgan banker said the firm has rolled out AI tools in investment banking globally, signaling the tech has moved from desk-by-desk experimentation to a standardized operating layer. The key change is organizational: global deployment implies controls, training, and integration into core banker workflows rather than isolated copilots.

Action

Force a coverage/IB productivity plan that assumes peers are standardizing AI in pitchbooks, comps, and client prep—then measure cycle time and quality deltas desk-by-desk. Standardize approved tools and prompts, and lock down data handling (client confidentiality, MNPI) before “shadow AI” becomes the default banker stack.

Read article →
cybersecuritydive.com

NYDFS issues heightened cyber guidance explicitly calling out AI-amplified threats (deepfakes, social engineering) for financial firms

New York’s financial regulator (NYDFS) issued updated cyber threat guidance for banks, insurers, and other regulated financial firms, explicitly flagging the elevated threat environment and AI-enabled tactics. The thrust is operational readiness: stronger controls, faster detection/response, and executive-level attention to evolving attack methods.

Action

Run an AI-deepfake and “synthetic identity / synthetic employee” tabletop within 30 days (wire/ACH change-of-instructions, vendor payment, and privileged-access scenarios). Tighten identity proofing and payment verification (out-of-band callbacks, cryptographic signing where possible) because traditional voice/video cues are no longer reliable.

Read article →

General AI

Large language models & AI infrastructure

2 stories
lambda.ai

DeepSeek v4 lands quietly—another step-change in open-source capability banks can run inside their own walls

DeepSeek released v4 with relatively little hype, but it reinforces the trend that frontier-adjacent models are arriving in open form factors. For regulated enterprises, the practical change is procurement leverage: stronger non-US/non-Big-Tech options for private deployment keep improving.

Action

Reprice your internal GenAI platform roadmap assuming at least one open model is “good enough” for high-volume internal use cases (ops, call center summarization, knowledge search) and negotiate commercial model pricing accordingly. Stand up an evaluation harness (accuracy, toxicity, jailbreak, PII leakage, latency, cost) so you can swap models without rewiring workflows.

Read article →
blog.cloudflare.com

Cloudflare adds CASB monitoring for Anthropic via Claude Compliance API—practical control plane for enterprise rollout

Cloudflare extended its CASB to support Anthropic’s Claude Compliance API, letting security/compliance teams monitor Claude usage directly in Cloudflare without endpoint agents. This is a concrete step toward centralized visibility and policy enforcement as LLM usage spreads across the enterprise.

Action

Treat LLM tooling like SaaS: deploy centralized monitoring and DLP-style controls (who used what, where data went, what got generated) before usage scales further. Consolidate on one or two control-plane patterns (CASB + SIEM + audit logs) so model adoption doesn’t outrun your ability to evidence compliance.

Read article →

Get this in your inbox every morning

Free · No spam · Unsubscribe anytime

Subscribe free →