Insights &amp; News

OpenAI says Astra reaches the Critical cyber capability threshold

OpenAI says Astra now meets its Preparedness Framework Critical cyber capability threshold, with advanced access initially limited to testers and Daybreak Blue.

Anthropic launches Claude Fable 5.1 and Mythos 5.1 with different safety boundaries

Fable 5.1 is generally available while Mythos 5.1 uses trusted access for advanced cyber and life-sciences work, even though both share the same base model.

GitHub previews Copilot billing, unified-agent, and code-review policy changes

GitHub's September schedule reopens paid Business and Enterprise signups, unifies Copilot agent surfaces, changes chat retention, and makes Balanced the default review effort.

OpenAI connects Epic EHR and nine healthcare data sources to ChatGPT

ChatGPT for Healthcare adds an authorized Epic EHR integration and a public-data plugin covering nine official sources, with governed access for eligible users.

Anthropic hardens AI-agent evaluations after unauthorized actions

Anthropic describes new real-time classifiers, stronger sandboxes, paused high-risk RL environments, and partner controls after unauthorized actions in reduced-safeguard evaluations.

OpenAI study finds ChatGPT and critical-thinking training improve different parts of student work

In a randomized experiment with more than 1,000 Bocconi students, ChatGPT improved quality and coherence while causal-reasoning training produced more varied ideas; the effects complemented each other.

QwenCloud adds qwen-mt-image-2.0 for image translation across 55 languages

QwenCloud lists qwen-mt-image-2.0 in its August 28 model changelog, with 55-language image translation, layout preservation, terminology control, sensitive-word filtering, and product detection.

xAI updates Imagine API with five reference images, new ratios, and auto quality

xAI's August 28 release notes add up to five reference images, 21:9 and 5:2 aspect ratios, and an auto quality setting to grok-imagine-image-2.0.

Anthropic puts a built-in browser in Claude Cowork for delegated web tasks

Anthropic is rolling out a separate browser inside the Claude Cowork desktop app so agents can navigate pages and fill forms, while prompt-injection safeguards and access boundaries remain essential.

OpenAI plans to end its Cursor model contract after SpaceX acquisition

OpenAI says it plans to wind down its model-supply contract with Cursor after SpaceX's change of control, with a proposed November 12, 2026 shutoff date for the relationship—not Cursor as a whole.

Microsoft tests enterprise agents as Customer Zero shifts security from models to identity and controls

Microsoft describes a Customer Zero pilot in which roughly 100 internal users ran agent scenarios on Windows 365 Cloud PCs, combining Entra Agent IDs, Defender, Purview, Intune, and network controls.

Google Demand Gen update extends AI marketing from asset creation to video and conversation workflows

Google Ads says Multimodal Video Creation in Asset Studio is now generally available, while Demand Gen experiments add YouTube messaging conversations, travel personalization, and cross-format creative workflows.

OpenAI publishes first Jalapeño results as inference infrastructure targets interactive agents

OpenAI says its Jalapeño inference chip improved work per watt and end-to-end latency across several models and interactive workloads, although the results remain vendor-reported.

Salesforce and Anthropic launch Claudeforce to put Claude inside governed CRM workflows

Salesforce and Anthropic are connecting Claude to Salesforce data, business logic, permissions, and actions, starting with 37 prebuilt sales skills and governed agent tool calls.

Google launches Gemini Enterprise for Legal with governed agents for contracts, research, and compliance

Google Cloud has previewed Gemini Enterprise for Legal, combining legal skills, permission-aware connectors, agents, and a partner ecosystem in one governed workspace.

AWS and NVIDIA plan 2 million more GPUs for agentic and physical AI

AWS and NVIDIA plan to deploy 2 million additional GPUs across AWS infrastructure in 2027–2028, alongside CPU, memory, networking, open models, data software, and robotics tools.

Britive ARC reframes AI-agent access as task-scoped, zero-standing privilege

Britive has introduced ARC, a runtime authorization model that grants AI agents only the access required for a task, controls it while work runs, and revokes it when the task ends.

Aug 24, 2026

Caddi launches an agent that builds and governs back-office agents

Caddi has launched an AI agent that discovers repetitive work, learns processes through employee demonstrations, and builds auditable agents for professional-services back offices.

Aug 24, 2026

Google Cloud and Verizon put Gemini Enterprise across customer care, networks, and marketing

Google Cloud and Verizon have announced a strategic partnership spanning Gemini Enterprise, data infrastructure, network intelligence, customer experience, marketing orchestration, and employee productivity.

Aug 24, 2026