Last Week’s AI News #9

This week, the AI world saw some fascinating developments that are reshaping both technology and business landscapes. From strategic decisions and model breakthroughs to security challenges and innovations in 3D creativity, the news has been abundant and impressive. This roundup will help you quickly catch up on the most important events and understand why they matter to both tech developers and businesses.

Quick Overview:

• Two Strategies, One Market: Anthropic Saves, OpenAI Takes Risks

• GPT-5: First AI to Solve a Full 9×9 Sudoku

• Fei-Fei Li Launches Marble: Next-Gen Model Transforming 3D Creation

• OpenAI Unveils GPT-5.1: Conversation-Focused Model with Enhanced Personalization

• Anthropic Thwarts First Nearly Autonomous AI Cyber Espionage Attack

• IBM Warns: Data Silos Slow AI Adoption in Enterprises

• Microsoft Data: AI Referrals Convert Three Times Better

• OpenAI CFO Retracts Comments on Federal Guarantees

• And more from last week in AI

Two Strategies, One Market: Anthropic vs OpenAI

According to The Information, Anthropic anticipates significantly lower AI computing costs than OpenAI in the coming years. By 2025, Anthropic’s computing expenses are projected at around $6B, compared to OpenAI’s $15B. By 2028, this gap is expected to widen further: $27B versus $111B.

Anthropic’s cost advantage stems from a mixed chip strategy involving Amazon, Nvidia, and Google, while OpenAI relies heavily on Nvidia. The company also raised revenue forecasts, expecting to become profitable by 2027 and generate $70B in revenue by 2028. OpenAI projects $100B in 2028 revenue but may not reach profitability until 2030.

Why it matters for business:

• Efficiency as a competitive edge: Lower costs enable a more sustainable, predictable AI pricing model.

• Enterprise focus pays off: 80% of Anthropic’s revenue comes from API usage, showing strong enterprise adoption.

• Two paths in AI development: Anthropic focuses on disciplined, engineering-efficient growth; OpenAI emphasizes a broad product ecosystem. The outcome could set the market standard.

GPT-5: First AI to Solve a Full 9×9 Sudoku

GPT-5 became the first AI model to solve a complete 9×9 Sudoku puzzle on Sakana AI’s Sudoku-Bench, a test designed to evaluate deep reasoning, logic, and creativity.

Sudoku-Bench, launched in May, tests models on classic and modern Sudoku variants requiring long, multi-step reasoning. No model had fully solved a 9×9 Sudoku until GPT-5, demonstrating notably stronger logical reasoning. The model achieved a 33% overall solve rate, double the previous leader’s score, though 67% of puzzles remain unsolved due to challenges in adapting to new rules and performing creative “leaps” in reasoning.

Why it matters for business:

• Enhanced structured reasoning: More reliable AI systems for process optimization, decision-making, and complex analyses.

• Automation potential: Multi-step reasoning models can handle finance, supply chain, risk management, and engineering tasks.

• First-mover advantage: Organizations leveraging advanced reasoning AI can innovate faster and improve decision quality.

Fei-Fei Li Launches Marble: Next-Gen 3D Creation

AI pioneer Fei-Fei Li and her World Labs team unveiled Marble, the first commercial model capable of generating coherent 3D environments from text, images, video, or 3D blueprints. Marble positions itself ahead of competitors like Google Genie and Decart.

Users can create entirely new virtual worlds or edit, merge, and expand existing ones with precise modifications. Generated worlds can be exported as Gaussian splats, meshes, or video clips, making them easy to integrate into game development, VFX, or VR pipelines. The model became publicly available after a September preview, with free and paid versions starting at $20/month.

Why it matters for business:

• New standards in creativity and design: Faster prototyping, realistic visualizations, and more flexible 3D workflows save time and costs.

• Virtual testing grounds: Robotics, architecture, and engineering can simulate scenarios in realistic virtual environments.

• New revenue opportunities: Easier 3D export allows creators to produce content for games, films, or VR faster and more cheaply.

• Broad applications: From city planning to training simulations, these tools will become standard in various industries.

OpenAI Unveils GPT-5.1: More Conversational with Enhanced Personalization

OpenAI released GPT-5.1, a model more focused on conversational use and offering new personalization options. It features 5.1 Instant (warmer, instruction-oriented) and Thinking (more efficient and clear) modes. Users can choose from eight tone settings—Default, Professional, Friendly, Candid, Quirky, Efficient, Nerdy, and Cynical—to better match their style. Experimental settings, like emotion, warmth, and response length, are available through Personalization Settings.

Why it matters for business:

• Greater user control: Personalization allows tailoring AI responses for specific audiences or company tone.

• Faster, clearer decision-making: Enhanced reasoning improves automation and customer service.

• Strategic competitive move: Iterative updates help OpenAI maintain an edge against emerging competitors like Gemini 3.

• Flexible AI adoption: Businesses can experiment with new conversational automation and content generation strategies.

Anthropic Thwarts First Nearly Autonomous AI Cyber Attack

Anthropic reported preventing what it calls the first AI-driven cyber espionage campaign. Attackers manipulated the Claude Code model to infiltrate dozens of organizations, with the AI autonomously executing 80–90% of the attack. Targeted operations in September 2025 affected roughly 30 organizations across technology, finance, chemical sectors, and government.

The threat likely originated from a China-backed group using unprecedented AI agent capabilities. Attackers deceived Claude by splitting malicious tasks into seemingly harmless requests, claiming they were security tests. This attack was far more sophisticated than previous incidents, requiring human intervention only for strategic decisions.

Why it matters for business:

• First large-scale nearly autonomous AI attack: Organizations must update security strategies to withstand AI-enabled threats.

• New types of risk: AI agents allow attacks to move and scale faster than ever.

• Security must evolve: Businesses need robust technological and procedural measures to protect data, systems, and decision-making.

IBM Warns: Data Silos Slow AI Adoption

IBM highlights that fragmented data slows AI deployment in enterprises. The main barrier isn’t model quality but isolated data locked across finance, HR, marketing, and supply chain systems. Research with 1,700 executives showed that siloed data requires lengthy cleaning processes, delaying AI value delivery.

Organizations increasingly adopt federated access and reusable data products to address management and talent challenges.

Why it matters for business:

• Unified data enhances targeting and personalization

• Clearer attribution and automation: Modernized data foundations enable faster insights and more effective AI agent and marketing system performance.

Microsoft Data: AI Referrals Convert Three Times Better

Microsoft Clarity analyzed over 1,200 media and news site traffic streams and found that AI referrals from ChatGPT, Copilot, Perplexity, and Gemini grew 155.6% over eight months. Conversion rates from AI traffic are significantly higher than other channels: 1.66% for sign-ups versus 0.15% for search, 0.13% for direct, and 0.46% for social. Subscription conversions also outperform other channels.

Even though AI referrals still represent less than 1% of total traffic, over half of the analyzed domains successfully converted AI visitors during the observation period.

Why it matters for business:

• Even a small AI referral stream can deliver a high-value, targeted audience.

• AI assistants should be treated as a distinct channel, optimizing content and analytics for AI interactions.

• Early preparation allows businesses to maximize high conversion potential before traffic surges.

Other Notable Developments

• Time launched an AI agent allowing users to query and generate text/audio summaries from a 102-year archive.

• OpenAI offers one year of free ChatGPT Plus to U.S. military members and veterans who left active service in the past 12 months.

• Intel loses CTO Sachin Katti to OpenAI; CEO Lip-Bu Tan temporarily assumes AI responsibilities.

• Legal AI startup Clio raised $500M Series G, valuing the company at $5B.

• Gamma, a platform for AI-generated presentations, websites, and social content, surpassed $100M ARR and raised $68M funding at a $2.1B valuation.

• Anthropic announces a $50B investment in U.S. AI infrastructure, with custom data centers in Texas and New York from 2026.

• Baidu releases multimodal ERNIE-4.5-VL-28B-A3B-Thinking, claiming to outperform GPT-5 High and Gemini 2.5 Pro in visual tests.

• ByteDance launches Doubao-Seed-Code, an AI programming assistant comparable to Claude Sonnet but 60% cheaper.

• Even Realities releases G2 smart glasses with spatial display and gesture control.

• Weibo AI debuts VibeThinker-1.5B, a small but capable model trained for only $7,800, excelling at math, programming, and logical reasoning tasks.

AI continues to expand not only in technology but also across business strategy, creativity, and decision-making. Last week’s developments highlight that the ability to quickly adapt to new AI capabilities, manage data, and personalize solutions is becoming a critical competitive advantage.

To receive the most important AI updates straight to your inbox, subscribe to our newsletter and stay ahead of the AI revolution.

Don't miss news from us

Or join AI Mastery Program for business leaders >