Microsoft has launched Microsoft-Decision-1, an AI model designed for fast structured decisions rather than text generation. Announced by chairman Satya Nadella on October 9, 2026, the model scores options numerically in a single pass. Microsoft is already using it internally for incident response, quality control, and scientific discovery. It is available in Microsoft Foundry and through OpenRouter at $0.042 per million input tokens, with output tokens free.
In testing across 36 benchmarks with nearly 150,000 questions, Decision-1 proved about 35 times faster than other models. Xbox researchers used it to sort over 10,000 pieces of feedback, achieving results competitive with GPT-6 Sol at 14 times faster speed and one two-hundredth of the cost. The model rates options given a situation and fixed answer choices, making it well-suited for AI agents that need to judge their next move.
Meanwhile, Intel is entering the GPU market with its new D-Series chips to challenge Nvidia in AI data centers. The company is partnering with ASUS to roll out D-Series systems and expanding its AI hardware lineup with Gaudi accelerators and GPU Flex Series cards. These processors target AI agent tasks like sorting databases and loading applications. Intel's stock has risen 183% year to date on strong CPU demand.
On the infrastructure side, Western Digital reported strong fiscal fourth-quarter 2026 results driven by AI-driven data growth and cloud infrastructure needs. The company guided fiscal first-quarter 2027 revenue at about $4.1 billion, implying roughly 45% year-over-year growth. Management expects demand for higher-capacity drives including ePMR, UltraSMR, and future HAMR products to continue climbing, though softer margin guidance compared to peers remains a concern.
LA Sysco Technologies plays a supporting role in this hardware push, offering GPUs, DDR5 RAM, enterprise SSDs, HDDs, NICs, switches, and RAID products to help system integrators build customized AI infrastructure. By bundling networking components alongside compute and storage, the company simplifies procurement for large-scale projects. Separately, HPE and NVIDIA are set to discuss AI infrastructure at a Washington dinner on October 29, exploring trade-offs in cost, capability, and control for sovereign AI architecture.
AI-powered scamming has grown into a 422 billion dollar industry, with AI-generated scripts and stolen data fueling phishing attacks on social media. Fake websites mimicking banks and payment gateways are now sold as kits, making fraud faster and easier. Phishing pages on free developer platforms and AI app builders increased more than fourfold to 10,600 over two years. Scammers are also impersonating real journalists, such as KUOW reporter Katie Campbell, to run fake book club schemes targeting authors. At least 43 authors received the emails, and at least one person paid the impersonator.
On the defensive side, Starling Bank launched an AI-powered Scam Intelligence agent for its five million customers, detecting romance scams, investment fraud, deepfake phishing, and 10 other scam types. The feature was inspired by Cecilie Fjellhoy, who lost nearly £200,000 to a romance scam featured in the Netflix documentary The Tinder Swindler. UK Finance data showed Britons lost £576.4 million to fraud last year. The AI agent asks probing questions when a customer describes a suspicious situation and advises whether it is likely a scam.
Rein Security, a startup focused on protecting AI agents, raised $25 million in a Series A round led by Glilot Capital and Sienna Venture Capital. Since launching in January 2026, the company says its revenue grew eightfold and its customer base grew fivefold. Customers include Lemonade, Dun & Bradstreet, Flex, and Swimlane across healthcare, finance, and energy. The company uses patented sidecar technology to monitor and control AI agents at runtime.
In research, the CLEAT framework teaches AI agents to ask better clarifying questions by co-training a clarification-value controller and a task policy. Using Qwen3-4B, CLEAT achieved the highest mean score across all ten evaluation settings on four benchmarks, with gains up to 13.2 and 6.5 percentage points over the strongest baseline on UserGym Travel and IN3 benchmarks respectively. Separately, global AI investment trends suggest the cycle could redirect foreign capital toward Indian companies supplying power and equipment for AI infrastructure, even as India fell 19% in dollar terms this year.
Key Takeaways
- Microsoft launched Microsoft-Decision-1, a model for fast structured decisions, available at $0.042 per million input tokens through Microsoft Foundry and OpenRouter.
- Decision-1 is about 35 times faster than other models across 36 benchmarks, and Xbox researchers achieved results competitive with GPT-6 Sol at 14 times faster speed.
- Intel is entering the GPU market with D-Series chips to compete with Nvidia in AI data centers, partnering with ASUS and expanding its lineup with Gaudi accelerators.
- Western Digital guided fiscal Q1 2027 revenue at about $4.1 billion, implying roughly 45% year-over-year growth driven by AI storage demand.
- AI-powered scamming has grown into a $422 billion industry, with phishing pages on free platforms increasing more than fourfold to 10,600 over two years.
- Starling Bank rolled out an AI Scam Intelligence agent to its five million customers, inspired by a woman who lost nearly £200,000 to a romance scam.
- Rein Security raised $25 million in Series A funding to protect AI agents using patented sidecar technology, with revenue growing eightfold since January 2026.
- CLEAT framework teaches AI agents to ask better clarifying questions, achieving the highest mean score across ten evaluation settings on four benchmarks.
- Intel's stock has risen 183% year to date on strong CPU interest as the company expands into AI hardware.
- Global AI investment trends could redirect foreign capital toward Indian AI infrastructure suppliers after India fell 19% in dollar terms this year.
Microsoft launches Decision-1 AI model for fast structured decisions
Microsoft introduced Microsoft-Decision-1 on October 9, 2026, a model built for fast structured decisions instead of text generation. Chairman Satya Nadella announced it, and Microsoft is already using it internally for incident response, quality control, and scientific discovery. The model scores options with numbers in a single pass rather than writing paragraphs. It is available in Microsoft Foundry and through OpenRouter at $0.042 per million input tokens with output tokens free.
Microsoft Decision-1 helps AI agents judge their next move
Microsoft-Decision-1 is a small decision-scoring model that rates options given a situation and fixed answer choices. Microsoft tested it across 36 benchmarks with nearly 150,000 questions and found it about 35 times faster than other models. Xbox researchers used it to sort over 10,000 pieces of feedback, achieving results competitive with GPT-6 Sol at 14 times faster speed and one two-hundredth of the cost. The model is available through Microsoft Foundry and OpenRouter with pricing of $0.042 per million input tokens.
AI scammers impersonate KUOW reporter Katie Campbell in new fraud scheme
Scammers are using AI to impersonate KUOW reporter Katie Campbell, sending fake emails to authors claiming interest in their books for a non-existent KUOW Book Club. At least 43 authors received the emails, and at least one person paid the impersonator. The FBI's Internet Crime Complaint Center has noted that AI makes scams faster and more personalized. Campbell emphasizes that KUOW does not charge money for interviews or news coverage. The scam emails come from the email address k.campbelkuow.org@gmail.com.
Starling Bank launches AI scam detection tool for customers
Starling Bank created an AI-powered Scam Intelligence agent that detects romance scams, investment fraud, deepfake phishing, and 10 other scam types. The feature was inspired by Cecilie Fjellhoy, who lost nearly £200,000 to a romance scam and was featured in the Netflix documentary The Tinder Swindler. The bank rolled out the tool to its five million customers after UK Finance data showed Britons lost £576.4 million to fraud last year. The AI agent asks probing questions when a customer describes a suspicious situation and advises whether it is likely a scam.
CLEAT framework teaches AI agents to ask better clarifying questions
CLEAT is a new framework that co-trains a clarification-value controller and a task policy so AI agents can ask more useful questions. It measures how much value a question adds before answering and uses paired rollouts to estimate that value. Using Qwen3-4B, CLEAT achieved the highest mean score across all ten evaluation settings on four benchmarks. The framework showed gains up to 13.2 and 6.5 percentage points over the strongest baseline on UserGym Travel and IN3 benchmarks respectively.
Intel launches new D-Series GPU chips to compete with Nvidia in AI
Intel is entering the GPU market with its new D-Series chips to challenge Nvidia in AI data centers. The company is working with partners like ASUS to roll out D-Series systems. Intel is also expanding its AI hardware lineup with Gaudi accelerators and GPU Flex Series cards. The processors are designed to handle AI agent tasks like sorting databases and loading applications. Intel's stock has risen 183% year to date due to strong CPU interest.
Global AI investment shift could bring foreign money back to India
Market experts say the global AI investment cycle could redirect foreign capital toward Indian companies that supply power and equipment for AI infrastructure. In 2026, AI-focused investors favored markets like South Korea, Taiwan, Japan, and the US, while India fell 19% in dollar terms. Foreign institutional investors pulled over 4 lakh crore rupees from the Indian market this year. Experts believe Indian equities could attract foreign flows again as earnings improve and valuations become more competitive.
LA Sysco Technologies supplies AI hardware for system integrators
LA Sysco Technologies LLC offers a wide range of AI and server hardware to help system integrators build customized infrastructure. Its portfolio includes GPUs, DDR5 RAM, enterprise SSDs, HDDs, NICs, switches, HBAs, and RAID products. The company provides high-performance GPUs for AI server clusters and both SSD and HDD storage options to balance speed and cost. By offering networking components alongside compute and storage, LA Sysco simplifies hardware procurement for large-scale infrastructure projects.
Western Digital sees strong AI storage demand in fiscal 2026 results
Western Digital reported strong fiscal fourth-quarter 2026 results driven by demand for high-capacity storage and growing cloud infrastructure needs. The company guided fiscal first-quarter 2027 revenue at about 4.1 billion dollars, implying roughly 45% year-over-year growth. Management expects AI-driven data growth to fuel adoption of higher-capacity drives including ePMR, UltraSMR, and future HAMR products. However, softer margin guidance compared to peers remains a key risk for investors.
AI-powered scamming grows into a 422 billion dollar industry
Scamming has become a 422 billion dollar business with help from artificial intelligence tools. AI-generated scripts and stolen data help scammers target people on social media with phishing attacks. Fake websites mimicking banks and payment gateways are now sold as kits, making fraud faster and easier. Phishing pages found on free developer platforms and AI app builders increased more than fourfold to 10,600 over two years. Research shows more victims are choosing not to report scams at all.
Rein Security raises 25 million to protect AI agents
Rein Security, founded in 2024 and based in the US and Israel, raised USD 25 million in a Series A round led by Glilot Capital and Sienna Venture Capital. The company builds a runtime security platform that uses patented sidecar technology to monitor and control AI agents as they execute actions. Since launching in January 2026, Rein Security says its revenue grew eightfold and its customer base grew fivefold. Customers include Lemonade, Dun & Bradstreet, Flex, and Swimlane across industries like healthcare, finance, and energy.
HPE and NVIDIA to discuss AI infrastructure at Washington dinner
The Register is hosting a dinner in Washington DC on October 29 with HPE and NVIDIA to discuss which AI infrastructure options maintain real boundaries. Speakers include Bharath Ramesh and Tracy Mills from HPE and Bert Condensa from NVIDIA, with editor-in-chief Matt Rosoff leading the discussion. The event will explore trade-offs in cost, capability, and control for sovereign AI architecture. It is aimed at technology and risk leaders in healthcare, finance, government, and higher education who are making infrastructure decisions now.
Sources
- Microsoft Decision-1 Claims 35x Speed vs GPT-6 Sol
- Microsoft’s new model helps AI agents judge their next move
- Scammers used AI to impersonate me, a KUOW reporter. Here’s how to avoid such traps
- New AI feature detects romance scammers, investment heists and deepfake phishing attempts
- CLEAT: Learning Value-Guided Clarification for Interactive Agents
- Intel aims to boost AI turnaround with new GPU chips to take on Nvidia
- Global AI trade shift could bring foreign flows back to Indian market: Experts
- LA Sysco Technologies LLC: Best AI Hardware Supplier for System Integrators
- Is Western Digital’s (WDC) AI-Fueled Outlook Reframing Its Role in Cloud Infrastructure Leadership?
- Scamming Becomes $422 Billion Business With AI Assistance
- Rein Security secures USD 25 mln Series A
- SPONSORED: Which AI infrastructure option actually holds a boundary?
Comments
Please log in to post a comment.