Microsoft introduces MAI Code 2 as Google expands AI programs

Leopold Aschenbrenner’s Situational Awareness rebuilt its AI-related stock portfolio in the third quarter after selling public positions in July, when its assets had fallen from a peak of $45 billion to about $10 billion, mostly in private investments. Filings showed stakes in SharonAI, CleanSpark, IREN, Bitdeer Technologies and Fermi, including nearly $39 million in CleanSpark. The fund also reported call options tied to companies and funds including Nebius, Sandisk, Bloom Energy, CoreWeave and the Roundhill Memory ETF. Its holdings after September are unclear. The portfolio faced October losses: SharonAI fell 58%, IREN 41% and Nebius 35%; the disclosures followed reports that the fund lost 67% of its portfolio value in July.

Island researchers found 186 misspelled domains impersonating 27 AI brands, including ChatGPT, Claude, Perplexity, Midjourney and Cursor. The fake sites used ClickFix prompts to trick Windows users into running harmful PowerShell commands, offered browser hijackers through fake security checks, and promoted extensions that could alter search results and collect user data. Researchers linked the infrastructure to more than 200 domains targeting other sectors, including banks and telecom companies. As of November 3, 2026, 17 of 30 identified extensions remained in the Chrome Web Store. Island advises checking URLs carefully or using bookmarks.

At its Ignite conference in San Francisco, Microsoft introduced MAI Code 2, a lower-cost coding model designed for complex work across multiple files and software projects. The company said it nearly matches rival models while using less computing power. Microsoft also previewed a cybersecurity model for a limited group of customers and highlighted models for decision making, Microsoft IQ for organizing workplace knowledge, and Copilot Fusion, which combines multiple models. A model-selection feature is intended to balance output quality and cost. CEO Satya Nadella also acknowledged a regional outage the previous week and said Microsoft would work to improve.

Google is expanding AI education through two programs. In Japan, the Reskilling Consortium is offering the Japanese version of Google’s AI Professional Certificate free to the first 10,000 applicants; the 7-to-10-hour course covers Gemini, NotebookLM and Google AI Studio. Completers receive a certificate and can apply for three months of Google AI Pro. First Coursera login is required by March 31, 2027, when the consortium ends its activities. In India, Accenture LearnVantage and Google Cloud plan to provide AI training at more than 1,000 schools and colleges, including Gemini-powered labs and higher-education courses. Participating institutions and a rollout schedule have not been announced.

Databricks and Google are working on sharing lakehouse data without copying it, with attention to metadata. The report says Iceberg table labels such as “domain=finance” and column labels such as “pii=ssn” can travel with federated tables, supporting masking and helping AI agents find context. The authors do not detail how systems should resolve conflicting labels or specify cross-system write performance, latency and costs.

Other reports cover AI’s business and market effects. MSCI said trailing 12-month net subscription sales grew 26% year over year, supported by broader enterprise relationships and larger deals; it said AI is accelerating product innovation and freeing capacity for investment, while financial priorities remain unchanged. A research team introduced GainReason, which rewards recommendation-model reasoning according to how each sentence changes the likelihood of the target Semantic ID; tests on three public benchmarks showed gains over competing methods. PGY’s bank credit platform emphasizes repeatable lending products, funding access and disciplined AI use, with management targeting a 4%–5% margin by the middle of next year.

Grayscale attributed a sharp crypto sell-off partly to shrinking stablecoin supply and losses in AI-linked stocks, including Nvidia. Bitcoin fell about 17% in the first two weeks of November to $62,000, while a crypto-sector index dropped about 18%. Grayscale said stablecoin supply declines accelerated after the October 29 Binance cyberattack; Bitcoin’s fall below its 200-day moving average and leveraged-position liquidations added selling pressure. The firm described the market as oversold but cautioned that this does not mean prices will recover soon.

Key Takeaways

  • Situational Awareness rebuilt its third-quarter AI stock portfolio, including nearly $39 million in CleanSpark, before steep October losses in SharonAI, IREN and Nebius.
  • Researchers identified 186 typo-squatting domains imitating 27 AI brands, including ChatGPT and Midjourney; they warn users to verify URLs and report that 17 identified extensions remained in the Chrome Web Store on November 3, 2026.
  • Microsoft introduced MAI Code 2, a lower-cost coding model, and previewed cybersecurity and decision-making models alongside Microsoft IQ and Copilot Fusion.
  • Japan’s Reskilling Consortium offers Google’s AI Professional Certificate free to its first 10,000 applicants; eligible completers can apply for three months of Google AI Pro.
  • Accenture and Google Cloud plan AI education at more than 1,000 Indian schools and colleges, with Gemini-powered labs and courses, but have not announced institutions or timing.
  • Databricks and Google are exploring how Iceberg table and column labels can remain attached to federated lakehouse data without copying it.
  • MSCI reported 26% year-over-year growth in trailing 12-month net subscription sales and cited enterprise deals and AI-enabled product development.
  • GainReason uses changes in target Semantic ID likelihood to train recommendation reasoning and outperformed competing methods in tests on three public benchmarks.
  • PGY management cited a 4%–5% margin target by the middle of next year and emphasized repeatable lending products, funding access and disciplined AI use.
  • Grayscale linked November’s crypto downturn to shrinking stablecoin supply, AI-stock weakness including Nvidia, and leveraged liquidations; it said an oversold market does not guarantee a near-term recovery.

Situational Awareness rebuilds its AI portfolio after summer selloff

Situational Awareness disclosed new investments in its filing for the quarter ending in September. The fund held SharonAI, CleanSpark, IREN, Bitdeer Technologies, and Fermi, with SharonAI as its largest holding. It also owned call options for Nebius, Sandisk, the Roundhill Memory ETF, and Bloom Energy. The fund had sold its public stock positions in July after its assets fell from a peak of $45 billion to about $10 billion, mostly in private investments. AI and chip stocks later fell sharply, and the fund’s latest positions after September are unknown.

Situational Awareness rebuilt its AI stock portfolio before October losses

Leopold Aschenbrenner’s Situational Awareness hedge fund rebuilt its AI stock portfolio in the third quarter. It took major stakes in SharonAI, CleanSpark, IREN, Bitdeer Technologies, and Fermi. The fund’s CleanSpark stake was nearly $39 million. These holdings faced sharp losses in recent weeks, according to the article.

AI fund rebuilt its holdings before October’s market rout

Situational Awareness rebuilt its AI investments before an October slump. SharonAI shares fell 58% that month, IREN fell 41%, and Nebius lost 35%, according to the article. A fund filing showed call options in AI-linked stocks, including Bloom Energy, CoreWeave, and the Roundhill Memory ETF. Investors were concerned that Democratic gains in the midterm elections could affect data center development and that technology companies might be relying too much on debt to fund AI expansion. The article also says AI-related stocks remained under pressure.

Situational Awareness adds AI bets after a reported 67% portfolio loss

Leopold Aschenbrenner’s Situational Awareness held AI infrastructure companies including Nebius, Sandisk, and Bloom Energy in the third quarter. A filing showed call options in those companies and the Roundhill Memory ETF as of September 30, but did not say whether the fund kept or sold them during October’s market selloff. The fund also held about $39 million in CleanSpark, more than $21 million in IREN, over $20 million in Bitdeer, and more than $11 million in Fermi. SharonAI, IREN, and Nebius fell sharply in October. The new disclosure followed reports that the fund lost 67% of its portfolio value in July.

Leopold Aschenbrenner’s fund returns to AI investments after summer crash

Leopold Aschenbrenner’s hedge fund, Situational Awareness, added AI investments again in the third quarter after a summer market crash. The article says the fund piled back into AI bets, but does not provide details about its holdings or the size of its positions.

Typoed AI website addresses can lead users to malware

Island researchers found 186 misspelled website domains imitating 27 AI brands, including ChatGPT, Claude, Perplexity, Midjourney, and Cursor. The fake sites used attacks that could trick Windows users into running harmful PowerShell commands, install browser hijackers, or push malicious browser extensions. Researchers linked the operation’s infrastructure to 211 entry domains, including lookalikes of banks, insurers, telecom companies, and enterprise software providers. Some domains may still be active, and 17 of 30 identified extensions remained in the Chrome Web Store as of November 3, 2026. Island advises users to check web addresses carefully or use bookmarks.

Fake AI websites use misspelled domains to spread malware

Researchers identified 186 misspelled domains that imitate 27 AI brands, including ChatGPT, Claude, and Perplexity. The sites use several tactics, including ClickFix attacks that urge visitors to run harmful PowerShell commands, fake security checks that offer a browser hijacker, and malicious extensions that change search results and collect user data. Some extensions have been removed from the Chrome Web Store, but researchers say many fake domains may remain active. They linked the campaign’s infrastructure to more than 200 domains targeting banks, telecom firms, and enterprise software providers. Experts advise checking web addresses or using bookmarks to reach legitimate sites.

Microsoft introduces lower-cost AI models for coding and cybersecurity

At its Ignite conference in San Francisco, Microsoft introduced MAI Code 2, a lower-cost model for programming, and previewed a cybersecurity model for a small group of customers. Microsoft says MAI Code 2 nearly matches rival coding models while using less computing power, which can lower the cost of running it. The cybersecurity model aims to help teams find and handle software weaknesses. Microsoft plans to offer specialized models through products such as GitHub and its security tools, with controls for their use.

Microsoft highlights lower AI costs and task-focused models at Ignite

At its flagship Ignite conference, Microsoft focused on controlling AI costs as businesses use more tokens. AI CEO Mustafa Suleyman said companies want strong price performance and models that feel tailored to them. CEO Satya Nadella highlighted MAI Code 2, designed for complex programming tasks across multiple files and entire software projects, along with new models for cybersecurity and decision making. Microsoft also introduced Microsoft IQ to help businesses organize and share employees’ knowledge, expertise, and tools. Copilot Fusion combines multiple models, while a model selection feature aims to balance output quality and cost. Nadella also acknowledged a regional outage the previous week and said Microsoft would work to improve.

MSCI reports growth in subscription sales and enterprise deals

MSCI said trailing 12-month net subscription sales rose 26% year over year. Broader relationships with enterprise customers and larger solution deals are supporting growth. The company said AI is speeding up product innovation and freeing capacity for investment. MSCI added that its financial priorities remain unchanged.

Japan offers 10,000 free Google AI certificates before consortium closes

The Japan Reskilling Consortium said on November 16 that it will offer the Japanese version of Google’s AI Professional Certificate course free to the first 10,000 applicants. The course normally costs about ¥7,700 and is open to new and current members. It takes an estimated 7 to 10 hours and teaches practical AI skills using Gemini, NotebookLM, and Google AI Studio; no prior experience is needed. People who complete the course receive a Google certificate and can apply for three months of Google AI Pro, normally ¥2,900 per month. Participants must log in to Coursera for the first time by March 31, 2027, before the consortium ends its activities that month.

GainReason uses predictive gains to improve recommendation reasoning

Researchers introduced GainReason, a framework for improving generative recommendation with large language models. It measures how each reasoning sentence changes the likelihood of the target Semantic ID, then uses those changes as more specific training rewards. The framework includes PG-based Reasoning Optimization and an auxiliary task called PG-based Reasoning Utilization, along with a custom causal attention mechanism. Tests on three public benchmarks found that GainReason consistently beat competing methods and improved performance.

Accenture and Google Cloud plan AI training at 1,000 Indian campuses

Accenture LearnVantage and Google Cloud plan to bring AI training to students and educators at more than 1,000 public and private schools and colleges in India. Accenture LearnVantage will design and deliver courses, while Google Cloud will provide technology through its Digital Campus offering. Plans include Gemini-powered school labs, guided lessons, personalised coaching, and higher education courses in areas such as generative AI, software development with AI, and AI engineering. Students who complete programmes and pass assessments can receive certificates. The companies have not yet named participating institutions or provided a rollout schedule.

Databricks and Google focus on sharing lakehouse data labels

A report describes Databricks and Google’s work to share lakehouse data without copying it, with a focus on metadata rather than storage. The authors said Iceberg table labels, such as domain=finance and status=certified, can travel with tables when they are federated. Column labels such as pii=ssn can also persist, which they said matters for masking and helping AI agents find context. The article says the authors did not explain how systems should handle conflicting labels or detail write performance, latency, and costs across systems.

PGY expands bank credit platform with repeatable products and disciplined AI

PGY’s platform helps banks grow consumer lending through underwriting technology, integrations, and funding. Management pointed to stable consumer credit and disciplined ramp-ups by partners. It also cited a target margin of 4% to 5% by the middle of next year. The article describes the company’s approach as focused on repeatable products, access to funding, and disciplined use of AI.

Grayscale links crypto sell-off to shrinking stablecoin supply and AI stock losses

Grayscale said a shrinking stablecoin supply and falling AI stocks helped drive a sharp cryptocurrency sell-off. Bitcoin fell about 17 percent in the first two weeks of November, reaching $62,000, while a crypto sector index lost about 18 percent. Grayscale said the stablecoin supply decline sped up after the October 29 cyberattack on Binance, and losses in AI-linked companies such as Nvidia may have pushed investors away from risky assets. Bitcoin dropping below its 200-day moving average and liquidations of leveraged positions added to selling pressure. Grayscale said the market is now oversold and that prices may interest medium- and long-term investors, but this does not mean prices will recover soon.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

AI investments AI stocks Situational Awareness CleanSpark AI cybersecurity typosquatting phishing Microsoft MAI Code 2 AI coding models Microsoft Copilot Google AI education AI skills training Gemini Databricks Google Cloud Apache Iceberg data federation AI recommendation systems MSCI cryptocurrency market

Comments

Loading...