The Office of the Comptroller of the Currency released new guidance aimed at how bank examiners should evaluate artificial intelligence use within financial institutions. The framework, issued in early November, tells examiners to monitor AI adoption but not to treat it as a separate risk category. Banks are not required to create new risk management rules specifically for AI or define a separate AI risk appetite under the guidance.
Instead, the OCC framework supplements existing risk management practices and directs examiners to focus on what could happen if AI systems make errors. Higher-risk AI applications may need tighter limits or more human oversight, but the guidance itself is not legally binding. Still, banks typically follow OCC recommendations closely.
In Texas, Democratic wins in key statewide races could slow the rapid expansion of AI data centers. Candidates who support stricter regulation of data centers now control the governor's office and a Senate seat. Texas was expected to meet nearly 30 percent of national AI electricity demand by 2028, but incoming leaders like Gina Hinojosa are calling for a moratorium on new data center projects. Large developments such as Stargate Frontier and Nexus Hubbard are already under construction, and local officials and unions may resist sweeping changes.
Apple is tightening its spending on AI tools for employees. In October, the company reduced some engineers' daily AI budget from $300 to $200, with plans to possibly lower it further to $100. Some teams face limits as low as $25 per day or are being asked to use cheaper models. Apple employees can still access tools like Claude, ChatGPT, and Gemini, but the cost cuts come shortly after CEO John Ternus took over in September and have raised concerns about possible restructuring.
Block's stock rose on strong results tied to Jack Dorsey's AI-driven strategy. Gross profit climbed 22 percent year over year to $3.2 billion, and adjusted operating income jumped 92 percent to $923 million. Cash App reached 60 million monthly transacting users in September. Block also plans to test its Buzz AI agent platform with customers and expects to generate first revenue from AI tools in the fourth quarter.
SoundHound AI partnered with LG Electronics to build a commerce channel on roughly 20 million LG Smart TVs in the United States. The service will begin with restaurant ordering and later expand to retail and other merchants. It will draw on SoundHound AI's network of more than 15,000 restaurant locations and its transaction routing and checkout platform. The deal could reshape how investors view SoundHound AI's growth potential.
Adani Enterprises and Jabil announced a proposed 50:50 joint venture to develop and manufacture AI and data-center hardware. The partnership will focus on servers, storage systems, and networking equipment, and will also explore edge computing, 5G, and IoT technologies. Shares of Adani Enterprises rose 10 percent after the announcement, and the venture is expected to create jobs and boost regional economic growth.
Researchers introduced a method called Forgetting to Improve that removes harmful data from machine learning models. The approach uses an influence function for Gaussian process models to identify and cut data that hurts performance on a specific task. It targets mismatches between models and real-world data and showed strong results in Bayesian optimization, model-based reinforcement learning, and transductive learning.
A separate study on user-centered agentic recommendation found that giving users access to more competing platforms does not always improve their welfare. The research, which used synthetic data and Amazon-based markets, showed that credibility assessment can help users benefit when markets and mechanisms meet certain conditions. Platforms can change item descriptions without changing item value, and users judge credibility based on feedback.
Precursor Ventures is offering a paid part-time internship focused on auditing and improving AI tools used across the firm. Interns will map the AI tools in use, build a shared dashboard, and reduce duplicated work. The role requires about 15 hours per week and is open to US students authorized to work, with a deadline of November 29, 2026.
In a technical advance, researchers introduced the Cone Alignment Index, a method for measuring sparsity in Transformer attention mechanisms using Lorentz-cone geometry. The approach presents two projections that create structured sparsity patterns while preserving the same active support, and is designed to accelerate sparse attention in Transformers.
Key Takeaways
- The OCC released guidance telling bank examiners to monitor AI use but not treat it as a separate risk category, with higher-risk uses needing tighter oversight.
- Democratic wins in Texas raise the chance of stricter rules and possible moratoriums on AI data center expansion in a state expected to meet nearly 30 percent of national AI electricity demand by 2028.
- Apple cut some engineers' daily AI spending from $300 to $200, with plans to lower it further, while letting employees use Claude, ChatGPT, and Gemini.
- Block reported gross profit of $3.2 billion, up 22 percent year over year, and expects first revenue from its Buzz AI agent platform in the fourth quarter.
- SoundHound AI partnered with LG to build a commerce channel on about 20 million LG Smart TVs in the U.S., starting with restaurant ordering.
- Adani Enterprises and Jabil announced a 50:50 joint venture to develop AI and data-center hardware including servers, storage, and networking equipment.
- The Forgetting to Improve method uses influence functions to remove harmful data from machine learning models, showing strong results in multiple domains.
- A study on agentic recommendation found that more platform access does not always improve user welfare, using synthetic data and Amazon-based markets.
- Precursor Ventures is hiring interns to audit AI tools across the firm, requiring about 15 hours per week with a deadline of November 29, 2026.
- The Cone Alignment Index offers a new way to measure and accelerate sparse attention in Transformer models using Lorentz-cone geometry.
OCC issues first guidance on how bank examiners should review AI use
The Office of the Comptroller of the Currency released new guidance on November 4 for bank examiners. It tells examiners to watch how banks use AI but not to treat AI as a separate risk. Banks do not need to create new risk rules just for AI. The guidance is not legally binding but banks usually follow it closely.
OCC tells bank supervisors to monitor AI without treating it as unique risk
New guidance from the Office of the Comptroller of the Currency says examiners should keep an eye on banks' AI use. The guidance says AI should not be seen as a risk that is separate from other risks. Examiners should focus on what could happen if AI makes mistakes at a bank. Higher-risk AI uses may need tighter limits or more human oversight.
OCC releases framework to guide examiners evaluating bank AI use
The OCC released a supervisory framework on November 3 to help examiners review how banks use AI. The framework does not set enforceable rules or requirements for banks. It is meant to supplement existing risk management practices. Examiners should not treat AI as a separate risk category. Banks are not required to define a separate AI risk appetite under the framework.
Democratic wins in Texas may slow AI data center expansion
Democratic candidates won key races in Texas, raising the chance of stricter rules for AI data centers. Both the Texas Senate and governor's seat went to Democrats who support more data center regulation. Texas was expected to meet nearly 30 percent of national AI electricity demand by 2028. Leaders like Gina Hinojosa now call for a moratorium on new data centers.
Texas Democrats' election wins threaten massive AI data center buildout
Democrats took control of key Texas offices and made data centers a major campaign issue. James Talarico won a Senate seat and Gina Hinojosa won the governor race. Large projects like Stargate Frontier and Nexus Hubbard are already under construction in the state. New leaders may push for delays or moratoriums, though local officials and unions could resist sweeping changes.
User-centered agentic recommendation and platform competition
Researchers proposed user-centered agentic recommendation, where user agents compare suggestions from competing platforms. Platforms can change item descriptions without changing item value, and users judge credibility from feedback. The study finds that more platform access does not always improve user welfare. Credibility assessment can help users benefit when markets and mechanisms meet certain conditions. Tests used synthetic data and Amazon-based markets.
Forgetting to Improve removes harmful data in active learning
The paper introduces Forgetting to Improve, a method that removes harmful data from machine learning models. It uses an influence function for Gaussian process models to cut data that hurts a specific task. The approach targets mismatches between models and real-world data. The method showed strong results in Bayesian optimization, model-based reinforcement learning, and transductive learning.
SoundHound AI partners with LG on smart TV commerce
SoundHound AI partnered with LG Electronics to build a commerce channel on about 20 million LG Smart TVs in the U.S. The service will start with restaurant ordering and later expand to retail and other merchants. It will use SoundHound AI's network of more than 15,000 restaurant locations and its transaction routing and checkout platform. The deal may change how investors view SoundHound AI's growth story.
Block stock rises as Jack Dorsey's AI strategy gains support
Block shares jumped after strong results tied to Jack Dorsey's AI-driven workforce cuts. Gross profit rose 22% year over year to $3.2 billion, and adjusted operating income rose 92% to $923 million. Cash App reached 60 million monthly transacting users in September. Block also plans to test its Buzz AI agent platform with customers and expects first revenue from AI tools in the fourth quarter.
Adani Enterprises and Jabil plan AI hardware joint venture
Adani Enterprises announced a proposed 50:50 joint venture with Jabil to develop and make AI and data-center hardware. The partnership will focus on servers, storage systems, and networking equipment. It will also explore edge computing, 5G, and IoT technologies. Shares of Adani Enterprises rose 10% after the announcement. The venture is expected to create jobs and boost regional economic growth.
Precursor Ventures pays interns to audit and improve AI tools
Precursor Ventures is offering a paid part-time internship for 2027 focused on AI and investment operations. Interns will map the AI tools used across the firm, build a shared dashboard, and reduce duplicated work. The role requires about 15 hours per week and is open to US students authorized to work. Applicants must submit a link to something they built and explain what problem it solved. The deadline is November 29, 2026.
Apple cuts employee AI spending limits to save money
Apple is reducing how much it spends on AI tools for employees. In October, it cut some engineers' daily AI budget from $300 to $200, with plans to possibly lower it to $100. Some teams may be limited to just $25 per day or asked to use cheaper models. Apple lets employees use tools like Claude, ChatGPT, and Gemini. The moves come after CEO John Ternus took over on September 1 and have raised concerns about possible restructuring.
New method speeds up sparse attention in Transformers
This paper introduces the Cone Alignment Index (CAI), a way to measure sparsity using Lorentz-cone geometry. It also presents two projections called PLCP and PECP that create structured sparsity patterns. These methods preserve the same active support and can generate row-wise, column-wise, and diagonal masks. The approach is designed to accelerate Transformer attention mechanisms. The authors prove that the two projections differ only by a positive radial scaling factor.
Sources
- US bank regulator issues first-ever AI guidance for examiners
- US bank regulator issues first-ever AI guidance for examiners
- OCC Releases Supervisory Framework for Bank AI Use
- Democratic victories cloud outlook for AI data center rollout
- Texas was the new epicenter of the AI data center buildout. That might be kaput.
- Toward User-Centered Agentic Recommendation: When Does Platform Competition Benefit Users?
- Forgetting to Improve: Principled Data Removal in Active Learning
- How LG Smart TV Deal At SoundHound AI Stock Has Changed Its Investment Story
- Block stock jumps as Jack Dorsey's AI bet shows signs of paying off
- Adani Enterprises stock: Jabil MoU opens AI hardware path
- Precursor Ventures AI & Investment Operations Internship 2027
- Apple is Putting Employees on an AI Budget
- A Parametric Linear Conic Projection for Structured Sparse Attention
Comments
Please log in to post a comment.