Daily AI News | Nvidia introduces OpenShell controls for AI agents
# Daily AI News | Nvidia introduces OpenShell controls for AI agents
## Lead Story
### Nvidia introduces OpenShell controls for AI agents
Nvidia introduced OpenShell 0.1.0, an open runtime that limits the systems and data an agent can access. It combines sandboxed execution, controlled service access, credential management and policy analysis, enforcing permissions outside the agent workload.
**AI take:** Useful autonomy requires enforceable access limits.
## More News
### Meta starts an enterprise AI platform business
Meta announced Meta Enterprise Platform as a new business. Its initial plan is to bring technologies including Muse, Meta Business Agent, Muse API and Muse Code to businesses and developers.
**AI take:** Availability of individual products will matter to buyers.
### MiniMax Code adds M3.1-Flash-Preview and resets quotas
MiniMax announced that M3.1-Flash-Preview is available in MiniMax Code. It also issued reset cards for all users’ Token Plan quotas, alongside check-in credits and free-token promotions.
**AI take:** The quota reset creates room to evaluate the model.
### Kling 4.0 Flash begins a limited preview
Kling 4.0 Flash opened a limited preview for Black Gold annual members on September 28. The full Kling 4.0 release is planned for October; current access is a restricted Flash preview.
**AI take:** Production plans should follow the version actually available.
### H Company introduces the Holo4 agent model family
H Company introduced Holo4 in two sizes: a 27B dense model and a 35B-A3B mixture-of-experts model. Both are available through the H Models API, alongside the updated Holotron4 Nano.
**AI take:** Different sizes offer more deployment and cost options.
### Base Code connects GitHub repositories to cloud AI collaboration
Base44’s Base Code connects existing GitHub repositories and lets an agent set up a cloud development environment. Team members can request changes in natural language, preview them and submit pull requests through existing review workflows.
**AI take:** Easier collaboration still needs sound code review.
### Gemini will start migrating Gems to skills on November 17
A notice in Gemini’s Gems manager says automatic migration to skills will begin on November 17, 2026. Users can continue using their Gems until those Gems migrate.
**AI take:** Existing setups remain usable; check their behavior after migration.
### Com2uS introduces the Hive Axyl game backend platform
Com2uS Platform introduced Hive Axyl at an AI game development event in Tokyo. Scheduled for release on September 30, the platform covers backend setup, feature integration and live operations.
**AI take:** Game prototypes still need backend and operating support.
### C1 releases its self-hosted AppHub application platform
C1 released AppHub as a public repository for deploying, authenticating and setting permissions for team-built applications. A skills file guides coding agents through building and launching apps, with the platform hosted in the team’s own cloud.
**AI take:** Internal apps need identity and permissions alongside deployment.
### Florida asks a court to restrict OpenAI model development
Florida’s attorney general asked a judge to bar OpenAI from developing new AI models without outside oversight, in a lawsuit alleging harm to children. This development is a request to the court; the report does not announce that it was granted.
**AI take:** The practical effect depends on the court’s decision.
### Anthropic reportedly will miss Australia’s October 1 AI hearing
Reuters reported that Anthropic will not attend an Australian Senate committee’s AI hearing on October 1. A person familiar with the matter said it sought another date because the invitation arrived late the previous week.
**AI take:** Attendance arrangements need to be checked for each committee.
### Nvidia adds $150 billion to its share repurchase authorization
Nvidia announced board approval for an additional $150 billion in share repurchases, bringing the remaining authorization to $235 billion. These figures describe authorized capacity, rather than completed purchases.
**AI take:** Capital allocation depends on how the authorization is executed.
### Samsung Electro-Mechanics plans a 6.78 trillion won substrate expansion
Samsung Electro-Mechanics plans to invest 6.78 trillion won in package substrate capacity in Sejong and Vietnam. Its Sejong site will increase the share of high-performance FC-BGA substrates for AI servers.
**AI take:** AI hardware growth also requires investment in chip packaging.
### Mobilint partners with Korean device makers on on-device AI chips
Mobilint announced participation in a Korean on-device AI semiconductor project with LG Electronics, Doosan Robotics and Daedong. The work targets robots and appliances, using product requirements from the device makers to guide development and validation.
**AI take:** On-device chips need validation against real device requirements.
### YCO and Aolani announce a Philippines GB300 cluster project
YCO Cloud and Aolani announced a project to deploy Nvidia GB300 NVL72 clusters in the Philippines. The initial phase is planned to exceed 10,000 Blackwell Ultra GPUs, with launch scheduled for the first quarter of 2027.
**AI take:** Planned capacity becomes useful only when deployment is delivered.
### KT Cloud joins the Gunsan AI data center project
KT Cloud announced participation in SGC Energy’s AI data center project in Gunsan, covering design, construction and operations. The first phase is planned at 60 megawatts, targeting operations in the first quarter of 2028.
**AI take:** Power and cooling will shape the project’s delivery capacity.
### Sharpa unveils a robot, dexterous hand and data glove at IROS
Sharpa unveiled the D01 tactile manipulation robot, W02 dexterous hand and AE01 exoskeleton data glove at IROS. The products address motion execution, fine hand manipulation, teleoperation and collection of manipulation data.
**AI take:** Their combined value needs validation on practical manipulation tasks.
### Nissan launches the UK ADventure autonomous driving research project
Nissan is leading the UK government-backed ADventure project with the University of Warwick’s WMG. Work in Cambridge will combine real-world testing and community engagement to study safe, acceptable and accessible autonomous passenger services.
**AI take:** Community acceptance is part of making autonomous services useful.
### Dubai opens the Mobility Labs autonomous vehicle testing facility
Dubai’s Roads and Transport Authority and Integrated Economic Zones Authority inaugurated Dubai Mobility Labs for autonomous and future mobility technologies. Its controlled facilities and public-road testing encompass 248 test scenarios.
**AI take:** Varied scenarios help expose problems in adapting to roads.
### Fedorov announces an Army of Robots defense technology initiative
Former Ukrainian defense minister Mykhailo Fedorov announced an Army of Robots initiative. The concept involves autonomous robots working with drones and unmanned ground systems; reporting describes it as an early-stage initiative.
**AI take:** Autonomous defense systems require clear accountability and reliability limits.
### Shinsegae I&C brings nine Meta smart-glasses models to Korea
Shinsegae I&C announced Korean distribution of Meta’s new AI glasses, including Meta-branded products and third-generation Ray-Ban Meta. The local launch covers nine models and twenty-one variants.
**AI take:** Local distribution changes access to purchases and related services.
### The Hey Ai PC launches a voice-operated Windows agent
Korean startup Hey Ai announced The Hey Ai PC for Windows. Users can make everyday-language requests for file searches, opening and saving files, web and app interaction, and mouse or keyboard actions.
**AI take:** Its practical value depends on reliable completion of everyday tasks.
### Gemini begins supporting Pixel Buds audio settings
An update to the Pixel Buds companion app adds the earbuds as a connected app in Gemini. Supported controls include equalizer presets, active noise cancellation, conversation detection and touch settings.
**AI take:** Common earbud settings become accessible through conversation.
### Princess Cruises launches a native ChatGPT planning app
Princess Cruises announced a native ChatGPT app for real-time cruise planning. The experience helps travelers discover, compare and plan trips while keeping them connected to Princess’s own booking ecosystem.
**AI take:** Travelers should still check the conditions at the booking stage.
### Insurance AI company Outmarket raises $34.5 million
Outmarket raised a $34.5 million Series B led by SignalFire, about four months after its previous round. The company uses AI to automate paperwork and administrative work for commercial insurance agencies and brokers.
**AI take:** Industry-specific products need to fit existing business workflows.
### Modulate secures funding for its voice analysis platform
Voice intelligence company Modulate secured a funding round led by Future Ventures, with Hyperplane and Lakestar participating. Its platform combines small models for enterprise functions including transcription, emotional analysis and deepfake detection.
**AI take:** Voice services need both content understanding and abuse detection.
## Rumors
### Report suggests Claude Sonnet 5.5 may be nearing release
TestingCatalog cited leaker Lyra as saying Anthropic partners received another Claude Sonnet 5.5 checkpoint and that release could be close. The report concerns partner testing and rumored timing; it does not establish a public launch.
**AI take:** Adoption plans should wait for official availability and pricing.
### Report says China may allow purchases of a new Nvidia chip
MT Newswires, citing The Information, reported that China may let some domestic companies buy Nvidia’s RTX Pro 5500 and has asked companies including Alibaba and ByteDance about purchase plans. Approval timing and criteria remain unclear in the report.
**AI take:** Procurement depends on clear policy and approval decisions.
