Skip to content
Daily Brief News Daily Brief

7 AI stories from June 2, 2026: Microsoft unveils 7 MAI models at Build, GitHub Copilot SDK goes GA in 6 languages, Anthropic expands Project Glasswing to 150 organizations, and 4 more product launches from OpenAI, GitHub, and the White House.

· by Pondero Newsdesk · 7 stories

AI news daily brief: 2026-06-03

Seven stories from June 2, 2026: four GitHub and Microsoft product launches tied to Build 2026, one Anthropic security program expansion, one OpenAI Codex usage milestone with new features, and one executive order from the White House on AI security.

Microsoft unveils 7 MAI models at Build 2026, including a reasoning model built without OpenAI data

At Build 2026 on June 2, 2026, Microsoft announced a family of seven in-house MAI models. The headliner is MAI-Thinking-1, a 35-billion-active-parameter Mixture-of-Experts reasoning model with a 256,000-token context window that scores 97.0% on AIME 2025, per CNBC's coverage. Microsoft said it was trained without any OpenAI data distillation, the first in-house reasoning model to meet that bar. MAI-Thinking-1 is entering private preview through Microsoft Foundry.

Also shipping on June 2 is MAI-Code-1-Flash, a 5-billion-parameter coding model now rolling out to all GitHub Copilot plans (Free, Pro, Pro+, and Max) starting with VS Code, per the GitHub Changelog. The full family also includes MAI-Image-2.5, MAI-Transcribe-1.5 (43 languages), and MAI-Voice-2 (15-plus languages).

Read more: MAI-Code-1-Flash is now available for GitHub Copilot

GitHub Copilot SDK reaches general availability in 6 languages with MCP and BYOK support

GitHub released the Copilot SDK as generally available on June 2, 2026. The SDK is available in Node.js/TypeScript, Python, Go, .NET, Rust, and Java, per the GitHub Changelog. It exposes the same agentic runtime that powers GitHub Copilot: planning, tool invocation, file edits, streaming, and multi-turn sessions.

Three features shipped new at GA: support for custom tools, Model Context Protocol (MCP) servers, and a bring-your-own-key (BYOK) option for users without a Copilot subscription. All existing Copilot subscribers, including Copilot Free, can use the SDK at no additional charge.

Read more: Copilot SDK is now generally available

Anthropic expands Project Glasswing to 150 organizations in 15-plus countries

Anthropic announced on June 2, 2026 that Project Glasswing, its collaborative vulnerability-scanning program, is expanding to approximately 150 new organizations across more than 15 countries. The expansion follows an initial April 2026 cohort of roughly 50 partners who found more than 10,000 high- or critical-severity security flaws using Claude Mythos Preview, per the Anthropic announcement.

The new organizations cover industries absent from the first cohort: power utilities, water systems, hospitals, telecom providers, and hardware vendors. Anthropic estimates a successful attack on most of these codebases could affect more than 100 million people. Each new partner must meet Anthropic security requirements before gaining access to Mythos capabilities.

Read more: Expanding Project Glasswing

OpenAI Codex passes 5 million weekly users and adds role-specific plugins, Sites, and annotations

More than 5 million people now use OpenAI Codex weekly, the company announced on June 2, 2026. Non-developers, including analysts, marketers, designers, researchers, investors, and bankers, make up about 20% of overall Codex users and are growing more than 3x as fast as developers, per OpenAI's announcement.

To serve those non-developer users, OpenAI launched six role-specific plugins bundling 62 apps and 110 skills. The Data Analytics plugin connects Snowflake, Databricks, Hex, and Tableau. Creative Production links Figma, Canva, and Shutterstock. Sales integrates Salesforce, HubSpot, and Slack. Two other features also shipped: Sites (in preview for Business and Enterprise plans), which publishes Codex output as a shareable interactive web app at a URL; and Annotations, which lets users flag and correct Codex output in context. Additional role plugins in Corporate Finance, Legal, and Strategy Consulting are listed as coming soon.

Read more: Codex for every role, tool, and workflow

GitHub Copilot cloud and local sandboxes enter public preview for isolated agent execution

GitHub shipped cloud and local sandboxes for the Copilot CLI in public preview on June 2, 2026, per the GitHub Changelog. Passing the cloud flag to the CLI provisions a remote sandbox in GitHub's infrastructure and runs agent tasks in an isolated environment, keeping the developer's local machine clean. Local sandboxes use Docker to containerize agent work on the developer's own hardware.

Both sandbox modes integrate with the Copilot Automations feature (covered below), so nightly jobs can run in sandboxes rather than on shared CI runners. Cloud sandboxes are available to Copilot Pro, Pro+, Business, and Enterprise subscribers. Sandbox compute-minute pricing is the next item to watch: cloud sandboxes are currently in public preview and billed under the usage-based billing system that went live June 1, 2026.

Read more: Cloud and local sandboxes for GitHub Copilot now in public preview

GitHub Copilot Automations let the cloud agent run on a schedule or trigger from repository events

GitHub shipped Copilot Automations on June 2, 2026, enabling the cloud agent to run on a schedule or in response to repository events without manual initiation, per the GitHub Changelog. Supported triggers include hourly, daily, and weekly schedules, plus issue creation and PR creation or update.

Documented use cases include automatically labeling incoming issues, running nightly failing-test detection with draft-PR fixes, and drafting release notes on a schedule. Automations are available for Copilot Pro, Pro+, Max, Business, and Enterprise plans. Public repository support is listed as coming soon.

Read more: Schedule and automate tasks with Copilot cloud agent

Trump signs voluntary AI security executive order with 30-day pre-release model review process

President Trump signed an executive order titled "Promoting Advanced Artificial Intelligence Innovation and Security" on June 2, 2026. The order invites AI companies to participate, on a voluntary basis, in a benchmarking process to determine whether a model qualifies as a "covered frontier model" with advanced cyber capabilities, per the White House action page. Companies that participate would provide government access up to 30 days before public release for those models.

The order also establishes a voluntary AI Cybersecurity Clearinghouse in coordination with industry and critical infrastructure operators. The text of the order explicitly states it does not create any mandatory licensing or pre-clearance requirement. Google's Kent Walker, OpenAI's Sam Altman, and Microsoft's Brad Smith all issued supportive statements, per CNBC's coverage. The key question over the coming months is whether the voluntary benchmarking process sees broad industry adoption and whether the "covered frontier model" designation hardens into a de facto classification standard.

Read more: Promoting Advanced Artificial Intelligence Innovation and Security

Sources