The sparse mixture-of-experts model is in limited early access while weights, a technical report, model card and developer artifacts remain promised for later in October.
Archive
Latest AI intelligence — Page 5
AI reporting, analysis, and product updates across every desk, ordered by publication time.
708 sourced postsThe 3,500-line framework connects deployed agent harnesses to reinforcement learning through an API proxy and Kubernetes control plane, with benchmark gains still based on Microsoft's own experiment.
The preview adds agent identity, field-level policy and audit controls between AI systems and enterprise APIs, with Intuit piloting the service.
The Webex plan joins managed and collaborative agents to enterprise context and customer workflows, but several central capabilities are still described as coming soon or later.
The new ChatGPT versions remain below critical capability thresholds but show several disclosed safety-evaluation regressions alongside stronger results in other tests.
The lightweight Anthropic model is entering GitHub's coding assistant and AWS's model platform, while rollout and performance claims remain vendor-controlled.
AWS's industrial predictive-maintenance service has reached its October 7 retirement date, moving existing customers toward IoT SiteWise or partner alternatives.
The open 740-million-parameter model puts text, code, images, video and audio into one retrieval space, while its efficiency and benchmark claims still come from Google's own launch testing.
The research preview can choose among single-model, escalation and cross-model critique workflows, but GitHub has not published deployment-level cost or reliability evidence in the launch update.
Gong Enrich and Assistant Deep Mode are generally available, while event-triggered Custom Agents, expanded dashboards and other Assistant features are still rolling out.
The release joins agent workflows to package policy, secrets, audit evidence and impact measurement, but availability ranges from generally available to beta, early access and future plans.
Lightwell Clearinghouse is now generally available for enterprise priority review and backported fixes, but the vendors did not identify the affected libraries or independent validation.
The proposed protocol separates an agent's verified identity from a customer's delegated authority, combining Agent2Agent discovery with OAuth-based permission grants.
The advisory body is reviewing selected government AI projects before deployment and examining agency responses after serious incidents, while agencies retain responsibility for decisions.
The supervisory framework covers every form of AI used by regulated firms, with governance expectations starting in 2027 and later requirements following in 2028.
An IDE-attribution gap left some agent use out of enterprise reports or assigned it to Copilot CLI, and the missing historical data cannot be recovered.
The new endpoint accepts text and images and returns typed answers, with a vendor-stated speed gain but sparse public detail about methodology, pricing or operating limits.
The research evaluation uses detailed criteria across 11 simulated tasks, but its vendor-reported scores and simulated times do not establish customer productivity or general legal reliability.
The disclosure pairs papers with revision protocols, reasoning summaries and machine-checkable formalizations, while independent validation and model access remain incomplete.
Executive agencies must use new terminology in non-statutory communications, while the underlying legal definition and historical records remain unchanged for now.
Paid subscribers and workspace users can now analyze recordings as files, while access, transcript accuracy and language performance remain variable.
API organizations now move through three named levels instead of five, making the progression simpler while leaving customers to verify model-specific limits in their own accounts.
The vendor has separated defensive, authorized red-team and specialized critical-system work, while eligibility, monitoring and effectiveness remain defined by Anthropic.
The agency has created a dedicated centre for embodied-AI development, while its first real operational humanoid remains a 2028 target rather than a deployed capability.