Microsoft is pushing hybrid AI directly into Windows PCs. The new MAI Code 1.1 Flash model runs locally with 137 billion parameters using 3-bit precision, helping balance speed and cost. GitHub Copilot is also getting local AI capabilities, and Copilot+ PCs will support on-device context, actions, and models. These features are expected to roll out over the coming months.
Security remains a major focus. Windows 11 pairs AI with Zero Trust principles, relying on TPM 2.0, Secure Boot, Virtualization-based security, and Microsoft Pluton on Copilot+ PCs. NPUs running over 40 TOPS allow IT teams to decide where AI workloads execute. Microsoft also launched Execution Containers (MXC), now generally available, which let developers define and enforce what resources an AI agent can access at runtime. Windows Entra separates agent activity from user activity, and the system extends Microsoft Agent 365 controls to local on-device agents.
On the business side, CAR Group reported strong FY26 results. Revenue reached A$1,253 million, up 6 percent, while adjusted EBITDA hit A$700 million. Adjusted NPAT rose to A$407 million, and net profit after tax climbed 14 percent to A$314 million. The company is investing in AI for personalized search and dealer tools across Australia, North America, Latin America, and Asia. FY27 guidance targets 11 to 14 percent constant-currency revenue growth and 10 to 13 percent adjusted EBITDA growth.
Intuit is also betting heavily on AI workflows. The company is embedding AI-driven tools across QuickBooks, tax, and accountant products. Analysts project revenue growth of about 8.9 percent annually, with earnings expected to reach $6.9 billion by 2029. A partnership with SimpleClosure aims to streamline payroll and account closure workflows, though success will depend on long-term customer adoption.
Other developments include Architect Financial Technologies launching Liquid Inference, a router that runs a live auction for every LLM request, where providers bid on each prompt and buyers pay the lowest qualifying offer. The first 500 users receive $20 of free inference. Meanwhile, Lumen Technologies partnered with Eminence Grey to offer sovereign AI infrastructure in the US, combining network capabilities with an AI platform to secure data and workloads for enterprise and government customers.
On the policy front, Capitol Hill lawmakers from both parties say they lack the political will to bail out AI companies if growth promises fall short. Senator Ruben Gallego warned AI firms expecting help should make other plans, while Senator Adam Schiff noted that a major AI collapse could put pension funds at risk. In a separate case, an American man was jailed for using about 10,000 bots to stream his own AI-generated songs, marking the first criminal conviction tied to AI music streaming fraud.
Key Takeaways
- Microsoft's MAI Code 1.1 Flash runs locally on Windows PCs with 137 billion parameters at 3-bit precision
- GitHub Copilot and Copilot+ PCs are getting local AI features to balance speed and cost
- Windows 11 combines AI with Zero Trust security, using NPUs over 40 TOPS for on-device workloads
- Microsoft Execution Containers (MXC) now available to enforce AI agent access policies at runtime
- CAR Group posted FY26 adjusted EBITDA of A$700 million and adjusted NPAT of A$407 million
- CAR Group's net profit after tax rose 14 percent to A$314 million with FY27 growth guidance of 11-14 percent
- Intuit projects earnings of $6.9 billion by 2029 driven by AI workflows across QuickBooks and tax products
- Architect Financial Technologies launches Liquid Inference with live auction pricing for LLM requests
- Lumen and Eminence Grey partner on US sovereign AI infrastructure for enterprise and government
- First criminal conviction for AI music streaming fraud as man jails for bot-streaming AI-generated songs
Microsoft brings hybrid AI to Windows PCs with new coding model
Microsoft announced a hybrid intelligence strategy for Windows PCs. The MAI Code 1.1 Flash model will run locally with 137 billion parameters using 3 bit precision. GitHub Copilot will also get local AI to balance speed and cost. Copilot+ PCs will gain local context, actions and models. Features are expected to roll out in coming months.
Windows 11 uses AI and Zero Trust to boost security
Windows 11 combines AI with Zero Trust security principles to support AI adoption. Features include TPM 2.0, Secure Boot, Virtualization-based security and Microsoft Pluton on Copilot+ PCs. On-device AI with NPUs over 40 TOPS lets IT teams choose where workloads run. Organizations can verify identity, device and data protections before scaling AI deployments.
Microsoft launches Execution Containers for AI agent security
Microsoft Execution Containers (MXC) are now generally available to manage AI agent security. MXC lets developers and IT teams define what resources an agent can access and enforce those policies at runtime. Windows Entra will distinguish agent activity from user activity. The system also extends Microsoft Agent 365 controls to local agents on-device.
CAR Group reports FY26 growth and AI product push
CAR Group reported FY26 revenue of A$1,253 million, up 6 percent. Adjusted EBITDA reached A$700 million and adjusted NPAT was A$407 million. The company grew across Australia, North America, Latin America and Asia. CAR Group invested in AI for personalised search and dealer tools. FY27 guidance targets 11 to 14 percent constant-currency revenue growth.
CAR Group sees FY27 expansion with AI and global growth
CAR Group reported net profit after tax up 14 percent to A$314 million for FY26. Adjusted NPAT was A$407 million and total dividend reached 86.0 cents per share. The company operates digital vehicle marketplaces across four regions. AI is being used to improve search, dealer software and transaction tools. FY27 guidance calls for 10 to 13 percent adjusted EBITDA growth.
Architect launches real-time auction for LLM inference
Architect Financial Technologies launched Liquid Inference, a router that runs a live auction for every LLM request. Providers bid on each prompt and the buyer pays the lowest offer that meets their rules. The max price is locked before generation begins. The first 500 users get $20 of free inference.
JIVEAdapter splits model updates into shared and task-specific parts
JIVEAdapter is a multi-task low-rank adapter that uses Joint and Individual Variation Explained. It splits every weight update into a shared Joint structure and a per-task Individual structure. The Individual structures are penalized to stay near-orthogonal to the Joint part. It was tested on GLUE and SuperGLUE with DeBERTaV3-base.
Learn2Play Bench tests how LLM agents learn from experience
Learn2Play Bench is a new benchmark using text-based games with novel or counterintuitive rules. It tests whether LLM agents can learn through interaction rather than relying on pretrained knowledge. The study found that keeping full records of actions and feedback helps learning. Top human players still outperform the evaluated agents. The agent harness also affects performance and cost.
Capitol Hill rejects idea of an AI industry bailout
Lawmakers from both parties say they lack the political will to bail out AI companies if the promised growth boom falls short. Senator Ruben Gallego warned AI firms expecting help should make other plans. Senator James Lankford said Congress does not need to manage this sector. Senator Adam Schiff noted a major AI collapse could risk pension funds.
Super Micro stock drops amid market pressure and profit-taking
Super Micro Computer stock plunged after a prior rally in the AI server sector. The decline came as the broader market faced pressure from rising Treasury yields and oil prices. The 10-year Treasury yield reached 5.29% and Brent crude moved above $104 a barrel. The company's trailing price-to-earnings ratio was about 13 times.
Intuit bets on AI workflows to drive growth and margins
Intuit is pushing AI driven tools across QuickBooks, tax, and accountant products to grow revenue and profits. Analysts expect revenue to grow about 8.9 percent yearly and earnings to reach 6.9 billion dollars by 2029. The SimpleClosure partnership aims to make payroll and account closure workflows smoother for businesses. Success depends on whether customers actually adopt these AI features over the long term.
Man jailed for using bots to stream AI generated songs
An American man became the first person imprisoned for AI assisted music streaming fraud. He commanded about 10,000 bots to stream his own AI generated songs. This tricked platforms into paying millions in fake royalty payments. The case marks the first criminal conviction tied to AI generated music streaming fraud.
Lumen and Eminence Grey launch US sovereign AI infrastructure
Lumen Technologies partnered with Eminence Grey to offer sovereign AI infrastructure in the US. The collaboration combines Lumen's network with Eminence Grey's AI platform to secure data and workloads. It aims to help enterprise, government, and AI customers run distributed AI applications across multiple sites. The companies expect the partnership to speed up AI adoption and support new services in the US market.
Sources
- Microsoft Takes AI Directly To PCs With New Coding Model And Security Tools
- Secure By Design: How Windows 11 Combines AI And Zero Trust To Drive AI Adoption
- Microsoft Execution Containers: Policy-driven containment for AI agents
- CAR Group (ASX:CAR): FY26 Growth and AI-Led Product Development Shape FY27 Focus
- CAR Group (ASX:CAR): FY26 Growth and AI-Led Product Development Shape FY27 Focus
- Architect Launches Liquid Inference, a Real-Time Auction for LLM Inference
- JIVEAdapter: A Multi-Task Additive Low-Rank Adapter via Joint and Individual Variation Explained (JIVE)
- Learn2Play Bench: How Well Do LLM Agents Learn from Experience in Unfamiliar Environments?
- AI bailout? Capitol Hill says dream on
- Super Micro Stock Plunges as AI Server Rally Hits a Speed Bump
- Will AI Workflow Adoption Change Intuit's (INTU) Narrative
- American jailed for commanding 10,000 bots to stream his own AI-generated songs and earn millions in fraudulent royalty payments, beating Taylor Swift — is the first person to end up in prison for AI-assisted music streaming crime
- Lumen, Eminence Grey provide sovereign AI infrastructure
Comments
Please log in to post a comment.