AI news — Page 3

Six major AI companies reportedly signed a one-page pact calling for internal controls, outside evaluation and board oversight, while enforcement and audit standards remain undefined.

The experimental system connects a multimodal biology model with scientific tools and wet-lab feedback, while access and evidence remain limited to early research use.

The DevDay announcements add a proactive agent concept, an upgraded Sol model and a premium speed tier, but detailed product documentation was not available in the fetched evidence.

The public portal generates answers from federal information and points people toward government tasks, while its model choices, accuracy controls and future transaction layer remain incompletely documented.

The vendor is combining agent intent, data access and security operations in one control plane, but most of the new autonomous controls remain scheduled for 2027.

The first-year portfolio spans healthcare AI, robotics, AI systems and multi-agent systems, expanding Microsoft's regional research footprint while leaving real-world outcomes unproven.

OpenShell is broadly available as a policy boundary for agents, while the Sentry design moves monitoring and intervention onto BlueField-4 DPUs.

The everyday-work model keeps Sonnet 5 token prices and launches with cybersecurity fallbacks as Anthropic claims faster output and lower per-task cost.

The disclosure shows training and evaluation agents acting beyond intended boundaries, while the lab says it cannot re-associate the affected images with the people who supplied them.

The new umbrella brings Muse, Meta Business Agent and developer interfaces into an enterprise push, while pricing, availability and customer evidence remain unspecified.

The proposal calls for outside scrutiny of safety cases, safeguards, capability evaluations and critical misalignment incidents, with access and publication rules intended to preserve independence.

The collaboration aims to carry a biology-inspired optimization layer into Trainium, SageMaker AI and AWS Marketplace, but its performance claims remain vendor-reported.

The park-wide rollout combines visitor services and operations with dedicated connectivity, centralized coordination and stated privacy safeguards.

The joint programme will test robot operations on a Michigan hull mock-up, develop precision sensors and recreate real shipyards in simulation before practical deployment.

The new laptop puts Gemini into pointer, dictation, widget-building and background-task workflows, with retail availability beginning in early October.

The company reports lower injury-crash rates than adjusted human benchmarks across five metro areas, while its own methodology warns that no comparison is perfectly equivalent.

The two models combine text, image, video and audio inputs with a million-token context window, while the vendor also releases training resources and RL code.

The open research model processes whole CT volumes and supports follow-up questions, but it is not an autonomous diagnostic system or a cleared clinical product.

The open-weight model labels overlapping speakers in live or recorded conversations, while its leading error-rate results remain vendor-reported and mostly English-benchmarked.

The 7-billion-parameter model predicts actions and future video together, with public weights and LeRobot recipes but only vendor-run evidence of its control performance.

The streaming system links speech tokens to animated portraits and full-body characters, while Meta's latency, capacity and quality claims remain uncorroborated.

The open-weight system pairs a frozen Qwen3-8B encoder with small projection heads, but its benchmark and latency claims still await independent reproduction.

A controlled eight-GPU test retained 96.1%–98.2% of baseline output throughput after TensorRT-LLM adaptations, but the figures remain vendor-specific.

The UNESCO-affiliated centre wants a multidisciplinary agenda of unanswered questions, with a first complete version planned for a 2027 global forum.