Autonomous systems, practical workflows and the permissions that matter.
42 stories
Apple says future Mac users will need more explicit approval before granting Full Disk Access, a permission that can expose files, mail, messages and browsing history. The company links the change to rising privacy risks from increasingly capable AI.
Meta has opened Muse Gadgets as an open-source project for builders. The SDKs support ESP32 and Raspberry Pi devices, with app pairing, tokens and a clear hardware-risk trade-off.
NVIDIA will sell a 64GB DGX Spark from partner systems starting Oct. 23 at $4,999. The company says it can run local agents on device and scale to two units with Sync Cluster Assistant, but 64GB remains better suited to inference than fine-tuning.
OpenAI’s Agents API can run a browser in an OpenAI-hosted environment, but each new website origin needs approval before access. The change matters for agent workflows, testing, research and compliance boundaries.
ElevenLabs says a $300 million employee tender valued it at $22 billion as enterprise voice agents now drive most of its growth, according to the company.
OpenAI is rolling out Dots as persistent ChatGPT agents for eligible paid users. The company describes read-only background research and action approvals; business buyers should verify those controls in a supervised pilot.
Reco says its new ServiceNow integration will inspect Otto agents, run more than 170 ServiceNow-specific checks, and route findings back into ServiceNow tickets and workflows. The operational appeal is tighter remediation; the caveat is that the core
Shopify says browser AI agents can now inspect and update checkouts, then submit orders only after buyer confirmation. The change extends its WebMCP stack from product discovery and carts into checkout, but human handoff still gates payment.
Instinct says it raised $1 billion at a reported $10 billion valuation, but the bigger question is whether a personal agent that can place calls and move between apps can scale safely.
Anthropic will miss a Senate AI hearing in Australia this week but is expected before another committee next week, as lawmakers respond to the OpenAI agent breach and weigh mandatory AI incident reporting.
NVIDIA has launched Open Agent Safety Platform, combining OpenShell and Sentry to enforce agent policy outside the model. The pitch is simple: safer AI agents need runtime and hardware boundaries, not just better prompts.
Meta says Muse can browse, fill forms and check out for users, but Amazon has already blocked it from shopping on its site. The bigger issue is who controls AI transactions.
OpenAI said its agents leaked 53 ChatGPT user images, but the deeper issue is control: once data is anonymized for training, the company says it may not be able to trace or notify the affected users.