Daily AI News | AWS open-sources the locally runnable Strands Decider 2B model
# Daily AI News | AWS open-sources the locally runnable Strands Decider 2B model
## Lead Story
### AWS open-sources the locally runnable Strands Decider 2B model
AWS Strands Labs has released Strands Decider 2B, an open-source decision model based on Qwen3.5-2B that is available now and can run locally. It selects from developer-defined options and returns confidence scores, supporting classification and next-step decisions in automated workflows.
**AI take:** A fixed set of answers makes the output easier for software to consume. Cost savings still need testing on the intended workload.
## More News
### ChatGPT adds virtual try-on and product favorites
OpenAI has globally launched two ChatGPT shopping features: virtual try-on and Favorites. Users can upload a selfie or full-body photo to preview clothing and accessories, and save products they like to the app’s Library for later reference.
**AI take:** Shopping conversations can now move from finding items to previewing a personal look, while favorites keep undecided options accessible.
### Google’s Suncatcher test satellite carries AI chips into orbit
On October 1, a SpaceX rocket sent the Suncatcher experimental satellite carrying Google TPUs into low-Earth orbit. The mission will evaluate how the AI chips handle launch and the space environment, and test a cooling system. It remains an experimental validation project.
**AI take:** The test addresses real orbital operating conditions. Commercial deployment still depends on further engineering and cost validation.
### Cloudflare launches Clef decision models on Workers AI
Cloudflare has released Clef and Clef-flash, two open-weight decision models now available on Workers AI. They target structured choices such as classification, request handling and task routing, and developers can also obtain the model weights.
**AI take:** Hosted access and downloadable weights offer two deployment paths for teams to evaluate against their own operating environment.
### Shopify introduces the conversational store builder Canvas
Shopify has introduced Canvas, allowing merchants to chat with Sidekick to edit store theme files and preview interactive results rendered from the actual code. The initial release is desktop-only and does not yet support third-party themes, app blocks or translations.
**AI take:** Merchants can describe changes directly, while existing stores still need to check theme and extension compatibility.
### Legato launches Frames hearing glasses from $999
Legato says its AI hearing glasses, Frames, are now available to buy from $999, targeting adults with up to moderate hearing loss. The hearing-assistance technology is built into the glasses’ arms, and the company says it enhances voices while reducing background noise.
**AI take:** The glasses incorporate hearing assistance into an everyday item. Price and the individual experience remain central to a purchase decision.
### Ray-Ban Meta Gen 3 glasses go on sale in India
Meta says Ray-Ban Meta Gen 3 glasses are now available in India from INR 44,300 through its websites and local electronics and eyewear retailers. The new generation adds a customizable action button for quick access to Meta AI.
**AI take:** Local availability and a published starting price give Indian buyers a clearer basis for considering the glasses.
### Barclays expands Claude across development and operations
Barclays is expanding its collaboration with Anthropic to use Claude in software development and banking operations. The rollout is expected to put Claude Code in the hands of about half its developers by the end of 2026 and most by the end of 2027. Those figures are targets.
**AI take:** A staged rollout makes adoption easier to track as the tools enter daily banking workflows. The coverage targets still need to be delivered.
### LG plans its first U.S. chiller factory in Virginia
LG has announced plans for its first U.S. chiller factory in Isle of Wight County, Virginia, to supply cooling equipment for AI data centers. Production is scheduled to begin in the first half of 2027, placing the project at the construction and planned-production stage.
**AI take:** Local manufacturing could add a supply source for data-center equipment. Actual deliveries depend on the factory entering production.
### Nebius acquires inference optimization company Inferize
Nebius has acquired Inferize and will bring its team and technology into the managed inference platform Token Factory. The company aims to shorten model startup delays, reduce idle GPUs and improve inference resource utilization. These are goals of the integration.
**AI take:** Inference platforms compete on scheduling and startup efficiency as well as capacity. The improvement must show up in customer workloads.
### Seoul joins Hyundai on a 500-bus autonomous transit plan
Seoul, Hyundai and the local bus association have signed an agreement to commercialize Level 4 autonomous city buses, targeting 500 vehicles by 2030. The plan would build on existing early-morning trials and gradually expand vehicles and operating hours.
**AI take:** Automated public transport requires coordination among manufacturers, the city and operators. Actual routes and capacity will show how deployment progresses.
### Armadin raises $255.5 million for security agent swarms
Cybersecurity company Armadin has raised $255.5 million in a Series B round co-led by a16z and Accel, at a valuation above $2.5 billion. Its platform uses AI agent swarms to identify weaknesses in corporate infrastructure, with assessments conducted in a sandbox according to the company.
**AI take:** Parallel agent investigations have gained fresh funding. Enterprises still need to assess detection results and operating boundaries in their own environments.
### Sharon AI secures a $356 million GPU-backed debt facility
Sharon AI has entered a committed $356 million senior secured debt facility backed by GPUs, priced at a fixed 9.95% interest rate excluding fees. The arrangement is intended to support AI infrastructure deployments backed by customer contracts.
**AI take:** Equipment financing can support compute expansion, while fixed interest creates an ongoing cost that project revenue must cover.
### Sidero Labs launches a private AI system and management platform
Sidero Labs has introduced Talos Linux for AI, an open-source immutable operating system, and Talos Cupar, a commercial platform for managing models, GPUs and private data on an organization’s own infrastructure. Cupar is currently in limited availability, with general availability planned for March 2027.
**AI take:** Separating the system foundation from the management layer helps organizations assess the operational scope of running their own AI platform.
### Meritz launches MOUM with Webull AI tools
Webull is powering Meritz Securities’ MOUM investment platform for Korean retail investors, integrating Vega AI, a global investor community and access to U.S. equities markets. Vega AI is designed to help users analyze company filings, earnings reports and news.
**AI take:** Information analysis and market access share one platform. Users still need to check the AI-generated analysis.
### EC-Council releases the ADG 2.0 AI governance framework
EC-Council has released ADG 2.0 with more than 180 controls in 12 families, mapped to over 90 regulations, standards and frameworks. It is offering the framework and crosswalks free to governments, regulators and standards bodies to support their governance work.
**AI take:** Crosswalks can reduce repetitive mapping work. Organizations adopting the framework still need to check the legal requirements that apply to them.
### Exabeam expands Nova AI’s persistent investigation capabilities
Exabeam has updated Nova AI on its New-Scale platform to collect context and entity profiles continuously and run additional searches during security investigations. Analysts supervise the work, and the company is also introducing AI updates for LogRhythm.
**AI take:** Maintaining investigation context can reduce the effort of switching among clues, while responsibility for decisions still needs to be clear.
### Satlyt raises $8 million for satellite computing software
Satlyt has raised an $8 million seed round to develop computing software for satellites. The company aims to let multiple satellites share large computing jobs, focusing on coordination and the allocation of work across satellites.
**AI take:** Orbital computing also needs software to organize resources across satellites. The funding supports that coordination layer.
### Agility and FORT partner on Digit 5 safety architecture
Agility Robotics and FORT Robotics have signed a memorandum of understanding to advance the safety architecture of the next-generation Digit 5 humanoid robot. The work covers a safety pendant, on-robot communications and interfaces to external safety systems, at the collaborative development stage.
**AI take:** Robots entering workplaces need to interact with site safety systems. Interfaces and communications are essential parts of that integration.
### Colosseum Copilot v2 updates project analysis and tool recommendations
Colosseum has launched Copilot v2 with expanded project analysis, tool tagging and recommendations tailored to developers’ technology stacks on Solana. It covers 8,286 projects across five hackathons and is free to access with a Colosseum account.
**AI take:** Using tool choices from earlier projects can help developers narrow the set of options they need to investigate for a new build.
### Linkhome launches public GPU Marketplace 2.0 rentals
Linkhome has launched GPU Marketplace 2.0 through Linkhome AI, allowing customers to rent GPU computing capacity supplied by participating individuals and data centers. The company disclosed the start of the public rental service in an October 1 Form 8-K filing.
**AI take:** The platform connects spare resources with compute demand. Its practical value will depend on available capacity and service performance.
### Photon raises $4.5 million for agents inside messaging apps
Photon has raised $4.5 million in seed funding to develop products that help developers build agents. Its tools target messaging apps including iMessage and WhatsApp, allowing developers to offer agent services through existing chat interfaces.
**AI take:** Using existing messaging interfaces can simplify access to a service. Developers still need to deliver useful tasks and clear interactions.
## Rumors
