AI startup Thinking Machines launches an open-weight AI model ↗
Mira Murati’s Thinking Machines unveiled Inkling, its first general-purpose model. The 975-billion-parameter system is open-weight, allowing developers to download, run and customise it instead of renting access through a closed API.
Inkling still trails the strongest proprietary models overall, but reportedly competes well on agent tasks. That matters - Western developers have been leaning surprisingly heavily on Chinese open models lately, and now there’s another heavyweight in the ring... sort of.
Apple Intelligence AI service registered with Chinese cyberspace regulator ↗
Apple Intelligence has cleared a major regulatory hurdle for use on iPhones in China. The registration opens the door to a long-delayed rollout, though no launch date was announced.
Apple’s local version will use capabilities from Alibaba’s Qwen alongside technology developed with Baidu. It’s an unusual AI sandwich - Apple hardware, Chinese models, and government approval holding the toothpick together.
Cadence rolls out AI agent to speed circuit board, chip packaging design ↗
Cadence launched AuraStack, an AI “super agent” that can design circuit boards and chip packages from plain-language instructions. It plans workflows, operates Cadence’s engineering tools and virtually tests the resulting designs.
The company says AuraStack can halve time to market and deliver productivity gains of up to 15 times on individual tasks. Nvidia, TSMC and Schneider Electric are among the early users - plainly, this is where agents stop arranging calendars and start touching the metal.
Anthropic, Blackstone bet the next trillion-dollar AI business is implementation, not just models ↗
Anthropic and a group of major financial backers have named their AI implementation venture Ode. The business has $1.5 billion behind it and aims to help large companies turn capable models into systems that function inside complicated organisations.
Ode grew from Fractional AI and will receive potential customers from its private-equity backers’ portfolios. The bet is fairly blunt: the next giant AI company might not build the smartest model - it might install the plumbing without flooding the office.
Microsoft is reportedly training salespeople to talk down OpenAI and Anthropic ↗
Microsoft reportedly instructed sales teams to contrast its own AI products with offerings from OpenAI, Google and Anthropic. Executives pushed an end-to-end systems pitch centred on efficiency, cost and tighter integration with Microsoft software.
One presentation reportedly characterised Claude as slower, less accurate and weaker on security inside Microsoft’s office apps. Awkward, yes - Microsoft is now openly competing with model providers it has relied upon for years. Partners, rivals, frenemies... pick Tuesday’s label.
Microsoft patches record number of security vulnerabilities, citing its use of AI ↗
Microsoft issued fixes for 570 security flaws across Windows, Office and other products, marking its largest monthly patch release. At least two were zero-day vulnerabilities already being exploited by attackers.
The company partly credited AI-assisted research with uncovering more dormant bugs. So AI is creating a bigger security headache while helping uncover old headaches too - peculiarly circular, like a smoke alarm learning to manufacture smoke.
Vint Cerf is working on a plan to unleash AI agents on the open internet ↗
Internet pioneer Vint Cerf joined Innovation Labs as an adviser on infrastructure for identifying AI agents online. The organisation’s DNSid proposal would connect agents to verified domain names and use cryptographic proofs to record their identities.
The idea addresses a fast-growing problem: autonomous agents may act across websites without a universal way to establish who controls them, what authority they hold or who is accountable. Basically passports for bots... except the passport might negotiate with another passport while you’re asleep.
FAQ
What is Inkling, and why does its debut matter?
Inkling is presented as a new launch, although the available details do not explain its product, market, or capabilities. A debut usually signals that a company or platform is moving beyond development and into public availability. The central questions are what problem Inkling solves, who it is built for, and how it differs from existing alternatives.
What does Apple clearing China AI mean for its rollout?
Apple clearing China AI suggests progress toward offering AI features in the Chinese market, where products often face local regulatory and operational requirements. This could remove a significant barrier to deployment, though approval by itself may not confirm an immediate launch. Availability, supported devices, local partnerships, and feature differences may still vary by region.
How does AuraStack automate chip development?
AuraStack is described as automating chips, which likely refers to streamlining parts of semiconductor design, configuration, testing, or deployment. Across many chip workflows, automation can reduce repetitive engineering tasks and improve consistency. Its practical impact depends on which stages AuraStack supports and how well it integrates with existing design and production tools.
What does Ode scaling its rollout mean for users?
Ode scaling its rollout indicates that access is expanding beyond an earlier or more limited release. Companies often broaden availability gradually so they can monitor reliability, demand, and user feedback before reaching a wider audience. Users should watch for updated eligibility, regional access, pricing, onboarding requirements, and any changes introduced as the rollout expands.
Why is Microsoft attacking rivals while patching 570 bugs?
The headline combines competitive positioning with a major security update. Microsoft appears to be criticising or challenging rivals while also addressing 570 software vulnerabilities or defects. For customers, the immediate priority is patch management: reviewing affected products, testing updates, and deploying critical fixes promptly in line with internal security procedures.
How does DNSid identify AI agents?
DNSid is described as a system for identifying agents, likely AI agents or automated services operating across networks. A common approach is to assign agents a verifiable identity that helps systems recognise, authenticate, or monitor them. This may support security, accountability, and access control, although the implementation details would determine how reliable and interoperable the identification process is.