-
Policy · TechCrunch · 2 Oct
TechCrunch reports Apple will add controls around macOS Full Disk Access after reports about AI apps reading private data. Apple said AI agents have increased the risks of this access and future grants will require very explicit user action.
Why it matters. Desktop agents on Mac may face stricter steps before getting access to files, messages, and browsing history.
SafetyAssistantsDevicesPermalinkRead at techcrunch.com
-
Launch · SiliconANGLE · 1 Oct
SiliconANGLE reports Microsoft released MAI-Transcribe-2-Streaming and two MAI-Voice text-to-speech models for voice agents. Microsoft said the streaming model updates transcripts over WebSocket, supports more than 60 languages and costs 54 cents per audio hour.
Why it matters. This gives developers Microsoft speech input and output models with more control over latency, language support and pricing for voice agents.
ModelsAssistantsVoicePermalinkRead at siliconangle.com
-
Launch · TechCrunch · 1 Oct
TechCrunch reports Shopify launched Canvas, a site builder that lets merchants create stores by chatting with its AI agent, Sidekick. Shopify said Canvas renders the store’s actual code in real time, but the initial launch lacks third-party themes, app blocks, markets, translations, rollouts, theme updates, and mobile support.
Why it matters. Canvas adds a chat-based way to change store code, with initial launch gaps in themes, app blocks, markets, translations, rollouts, theme updates, and desktop-only use.
AssistantsCompaniesPermalinkRead at techcrunch.com
-
Policy · InfoQ · 2 Oct
InfoQ reports Docker is taking Sandbox Kit Specification to the CNCF as an Apache 2.0 spec that packages an agent, its tools, and requested access in an OCI image. InfoQ says Docker Sandboxes is the first conforming runtime, and the posts do not say if CNCF has accepted it.
Why it matters. A standard OCI format for agent permissions could make requested access reviewable and portable, but enforcement still depends on Docker Sandboxes as the only conforming runtime.
Agents at workInfrastructurePermalinkRead at infoq.com
-
Analysis · The Verge · 2 Oct
The Verge reports OpenAI’s Dots work better on work tasks than personal errands, and the rollout starts with highest-tier accounts. The review found security checks and manual input often blocked shopping and account tasks, but Dot handled website edits and video assembly after getting access to the user’s computer.
Why it matters. For teams testing agent workflows, Dots may fit computer-based work better than consumer errands, but security checks and manual steps still limit end-to-end automation.
Agents at workAssistantsCompaniesPermalinkRead at theverge.com
-
Analysis · SiliconANGLE · 1 Oct
SiliconANGLE reports that NetApp and Nvidia are redesigning storage for AI factories by separating data and metadata so each can scale independently. NetApp's Arindam Banerjee said agentic workloads can bring thousands of concurrent data requests, pushing more API-driven, agent-friendly controls.
Why it matters. If your agents hit enterprise data at once, storage that scales metadata separately and exposes APIs can help keep GPUs utilized as workloads move into production.
Agents at workInfrastructureCompaniesPermalinkRead at siliconangle.com
-
Analysis · SiliconANGLE · 1 Oct
SiliconANGLE reports Nvidia says AI factory economics are shifting from individual GPUs to systems built for inference and tokens per watt. Ian Buck said power caps data centers, training continues alongside inference, and CoreWeave can help customers balance throughput with token speed.
Why it matters. For AI agent builders, the piece points to system design and efficiency, not just GPUs, as key to inference cost, latency and data center limits.
ModelsInfrastructureCompaniesPermalinkRead at siliconangle.com
-
Analysis · SiliconANGLE · 2 Oct
SiliconANGLE reports CrowdStrike is combining its security data with CoreWeave compute for SafeMind, and CoreWeave will help shape CrowdStrike's Cyber Superintelligence Lab. Bartley Richardson said CrowdStrike saw an eCrime breakout in 27 seconds in 2025, pushing machine-speed defense and trusted automation.
Why it matters. If you train and run models, this points to security tied to infrastructure, pipelines and automation that can respond faster and fix misconfigurations.
SafetyInfrastructureCompaniesPermalinkRead at siliconangle.com
-
Analysis · SiliconANGLE · 2 Oct
SiliconANGLE reports that IBM and CoreWeave are jointly engineering identity management and workload controls as agent testing adds tool and storage access to research systems. Brian Belgodere said IBM uses CoreWeave Sandboxes and single-tenant cluster design to isolate agent code.
Why it matters. For teams running agents, the report points to infrastructure choices around isolation, identity integration and security-performance tradeoffs as agent code moves beyond model training.
Agents at workInfrastructureCompaniesPermalinkRead at siliconangle.com
-
Policy · Axios · 3 Oct
Axios reports the U.S. Army plans an autonomous systems command. It also plans an acquisition executive to prioritize buying, after Defense Secretary Pete Hegseth announced robotic warfare projects called Meridian and Agincourt.
Why it matters. It suggests the U.S. Army may buy autonomous systems through a new command and acquisition executive.
Agents at workInfrastructurePermalinkRead at axios.com
-
Incident · WIRED · 2 Oct
WIRED reports nurses at HCA say the Palantir-powered Timpani scheduler can leave teams understaffed or exhausted and often misses shift requests. HCA says nursing leaders make the final schedules and the company is improving Timpani from nurses’ feedback.
Why it matters. For people using AI scheduling, complaints at HCA show how staffing, fatigue, and override decisions can become safety issues.
SafetyAgents at workCompaniesPermalinkRead at wired.com
-
Analysis · Wall Street Journal · 2 Oct
Wall Street Journal reports that TypeSafe CEO Diogo Almeida said Jev is used by about 25% of Fortune 500 companies. Almeida also said Jev reached a trillion tokens a day roughly a week earlier.
Why it matters. It suggests Jev already has notable enterprise use and scale, which matters if you build or run AI agents for big companies.
ModelsAgents at workCompaniesPermalinkRead at wsj.com