Daily AI News | Anthropic opens Claude Duet research preview for background tasks
# Daily AI News | Anthropic opens Claude Duet research preview for background tasks
## Lead Story
### Anthropic opens Claude Duet research preview for background tasks
Anthropic has opened a research preview of Claude Duet, an agent for professionals that can work in the background and return when it needs approval. Eligible US Pro and Max subscribers can join a waitlist, and usage counts against those plans' shared limits.
**AI take:** Background task execution moves assistants toward agent workflows, making permissions and user control central.
## More News
### Third-party pilot finds vulnerabilities in an OpenAI monitoring system
METR and Redwood Research said a pilot review of an internal OpenAI monitoring system found several vulnerabilities, including one that could keep hundreds of recent actions from being monitored. CNBC reported that OpenAI said it had patched issues affecting production safeguards.
**AI take:** The pilot underscores that monitoring systems also need independent testing; both findings and remediation remain attributed to the named parties.
### Uber plans to put about 30,000 employees on ChatGPT Work
Uber and OpenAI expanded their partnership. Uber told TechCrunch that after an earlier pilot with about 1,000 employees, it plans to roll ChatGPT Work out to roughly 30,000 people across its global workforce.
**AI take:** The move from a 1,000-person pilot to company-wide use shows enterprise AI adoption shifting toward deployment and governance at scale.
### MiniMax releases open weights for M3.1 Flash
MiniMax announced open weights for M3.1 Flash. Phoenix Tech reported 428 billion total parameters and 23 billion active parameters, along with integrations across several chip and inference platforms; the reported processing volume comes from MiniMax.
**AI take:** Open weights and multi-platform support give developers more deployment options, while performance still needs platform-specific evaluation.
### Canva releases its first image model, Canva Craft
Canva released Canva Craft, its first image-generation model, and made it available to users. Claims about its performance and cost advantages come from Canva.
**AI take:** Bringing image generation into a design platform lets users move from creation to layout within one workflow.
### 1Password adds agent logins without exposing passwords
1Password launched 1Password for Agents, which lets users authorize an AI agent to sign in to personal accounts without giving the agent the password. The service verifies the agent and asks the user to approve the login before filling in credentials.
**AI take:** The feature separates account authorization from handing over a password; its usefulness depends on which sites and agents it supports.
### NVIDIA launches Minerva operations platform for AI factories
NVIDIA launched the Minerva AIOps platform, which combines infrastructure and AI-workload telemetry with purpose-built models to help operators identify problems and keep AI factories running.
**AI take:** As AI infrastructure grows more complex, a unified view of hardware and workloads can shorten troubleshooting paths.
### Rivian acquires Third Dimension AI for driving simulation
Rivian announced that it acquired 3D spatial-generation company Third Dimension AI and plans to use its simulation platform to generate synthetic data for driver-assistance and self-driving development.
**AI take:** Bringing simulation tools in-house can help automakers create driving scenarios that are rare on public roads.
### KT Cloud plans a 50MW AI data center in Ansan
KT Cloud announced plans to build a 50MW AI data center in Ansan, South Korea, with completion targeted for 2029.
**AI take:** The planned capacity will not be available until construction is complete.
### Agentic AI Foundation adds four projects in its first year
The Agentic AI Foundation announced four hosted projects—Agent Skills, Open Responses, OpenEnv and OpenSandbox—bringing its first-year total to ten. The report attributes contributions to Anthropic, OpenAI, Hugging Face and Alibaba, respectively.
**AI take:** More shared projects can support reusable agent tools and test environments, though adoption will determine their impact.
### Agility Robotics selects NVIDIA IGX Thor for Digit 5
Agility Robotics said its Digit 5 humanoid robot will use NVIDIA IGX Thor for onboard computing as the companies expand their collaboration ahead of broader commercial deployment.
**AI take:** Commercial robotics depends on reliable onboard computing as well as the robot hardware itself.
### Meiro makes Local AI CDP generally available
Meiro announced general availability of Local AI CDP, with options including on-premises, private cloud and deployment in a customer's own cloud account. The company says the product is aimed at regulated industries and lets customers choose the underlying model.
**AI take:** Deployment choice can help organizations keep data and models in controlled environments, though compliance depends on each configuration.
### Qblox integrates NVIDIA DGX Spark with a quantum controller
Qblox said it integrated DGX Spark with its quantum system controller through NVIDIA NVQLink, reporting round-trip latency below 7.2 microseconds on DGX Spark. The performance figures are from company testing.
**AI take:** Connecting desktop GPUs with quantum-control equipment creates another path for hybrid-computing experiments.
### US Conec plans nearly $150 million fiber-optics expansion
Fiber-optics maker US Conec plans a roughly 200,000-square-foot plant in Hickory, North Carolina, with nearly $150 million in investment and 224 new jobs. State officials linked the expansion to rising data-center demand.
**AI take:** AI data-center growth is drawing investment into optical networking supply chains, beyond chips and servers.
### UK pensions regulator urges trustees to assess AI concentration risk
The UK's Pensions Regulator published guidance urging pension trustees to assess exposure to AI-linked companies and complex financing structures.
**AI take:** Index diversification may still hide concentration risk.
### AI pendant startup Nirva raises $25 million
Nirva announced a $25 million pre-Series A round after raising $8 million earlier in 2026. The company is developing an AI pendant.
**AI take:** The round signals investor interest in ambient-AI devices, though the product is still in development.
## Rumors
