BankingNewsAI Daily Brief ·
Trump unveils an AI task force
Banking AI
Financial institutions & fintech technology
New York regulator works with California and Illinois on AI
For the chief compliance officer at a US bank
New York’s work with California and Illinois raises the cost of treating AI governance as a single-state issue. Your bank needs one inventory and control standard for its own models and vendor AI, rather than separate fixes after state rules arrive.
New York’s Department of Financial Services is enforcing the Responsible Artificial Intelligence Safety and Education Act. The law requires major developers to register with the state’s financial regulator next month. Gov. Kathy Hochul signed the measure in December 2025. DFS Superintendent Kaitlin Asrow is leading the agency, one of the largest financial regulators in the US. The law requires registration as concerns about AI safety grow and companies discuss emergency shutdowns.
→ Action
AI governance: Inventory bank-built models and vendor AI systems, their owners, uses, and emergency shutdown controls.
Read article → from Bgov
General AI
Large language models & AI infrastructure
OpenAI tests visual ads in ChatGPT image generation
For the chief digital officer at a regional bank
ChatGPT is becoming a paid acquisition channel where bank ads can sit near sensitive consumer intent. Set rules for use cases, data sharing, attribution and exclusions before marketing buys access.
OpenAI will begin a US test later this month of visual ads during ChatGPT image generation with an initial group of advertisers. The ads will be labeled and separate from the image being created, and OpenAI says ads do not affect ChatGPT answers. ChatGPT reaches 1.2 billion people each week, according to OpenAI. Advertisers can send conversion data through integrations with Hightouch, Tealium and LiveRamp, while attribution partners support reporting, click attribution and Conversions API tools. Placement guardrails screen conversations for ad suitability, and qualifying advertisers can use Negative Phrases for their own placement requirements.
→ Action
Marketing governance: Require privacy and brand-safety approval before sending conversion data or buying ChatGPT ads.
Read article → from OpenAI
Trump unveils AI task force
For an AI-governance lead at a US bank
The White House review puts AI governance on a federal timetable without promising rules. Preserve proof of risk controls, third-party review and board oversight now; waiting for formal requirements could leave the bank unable to defend its practices.
President Donald Trump announced a White House “Super Intelligence Force” on AI, led by Director of National Intelligence Jay Clayton and the Federal Trade Commission chairman. The task force also includes the vice president, defense secretary, Treasury secretary and former AI czar David Sacks. It must complete a report within 120 days on AI risks and opportunities and on the federal government’s responsibility. Clayton told The Wall Street Journal that he expects mechanisms developed in dialogue with companies. Other officials said the task force will defer to firms on risk management while seeking to keep the US ahead of China.
→ Action
AI governance: Preserve evidence of model-risk controls, third-party reviews and board reporting for AI use.
Read article → from Semafor
Aleph Alpha releases 78B-parameter Kolibri AI model
For the CTO of a regional bank
Kolibri gives your bank an open-weight model for long German and English records without adding a closed-model vendor. Its open weights shift the burden to your team: prove performance, security and model controls before use.
Aleph Alpha released Kolibri, a 78.1-billion-parameter open-weight AI model with a context window of up to 1,048,576 tokens. The Heidelberg company built and trained the model in Europe, using infrastructure in Germany and Finland with 768 NVIDIA B200 GPUs. Kolibri is a Mixture-of-Experts transformer with about 3.46 billion active parameters per token, and it is tuned for English and German after training on 20 trillion to 24 trillion tokens. The model is available on Hugging Face under the Apache 2.0 license and requires about 78 GB of FP8 storage. Aleph Alpha published a 189-page technical report and reported a 96.9% score on AIME 2025.
→ Action
Model risk: Benchmark Kolibri on bank workloads and document accuracy, bias, security and human-oversight controls before any regulated deployment.
Read article → from Cryptobriefing
Get this in your inbox every morning
Free · No spam · Unsubscribe anytime