Claude Corps launches $150M, OpenAI acquires Ona, GitHub Agentic Workflows, Midjourney V8.1

The day of June 11, 2026 brings together four major announcements: Anthropic launches Claude Corps, a national fellowship program backed by $150 million; OpenAI acquires Ona to give Codex persistent cloud execution environments; GitHub opens its Agentic Workflows in public preview; and Midjourney makes V8.1 the default model with 4× resolution and 4-second generations.

Claude Corps — $150 million to train 1,000 fellows

June 11 — Anthropic is launching Claude Corps, a national fellowship program backed by an initial $150 million budget. The goal is concrete: train 1,000 young professionals to use Claude and place them in American nonprofits for one year, full time and in person.

Program structure:

Role	Organization	Responsibility
Funding & strategy	Anthropic	Claude expertise, unlimited tokens
Official employer	CodePath	Training, programming (1st US provider of university CS education)
Evaluation	Social Finance	Impact measurement, long-term financial vehicle

Conditions for fellows:

Full-time salary: $85,000/year + benefits
Duration: 12 months (initial intensive training + 5 h/week of ongoing training)
Criteria: 18+, less than 2 years of experience, work authorization in the United States

Cohort schedule:

Cohort	Applications until	Start
1st (100 fellows)	July 17, 2026	October 2026
2nd	Open	January 2027
3rd	Open	August 2027

Examples of host organizations: Braven (Chicago — first-generation student placement), Code the Dream (Durham — free coding training), International Rescue Committee, Code for America, YMCA of Greater Charlotte (300,000 people served).

Anthropic explicitly ties Claude Corps to its policy on the impact of AI on employment, presenting it as a model for expanding the benefits of AI during a period of major economic transformation. Part of the technical infrastructure will be open-sourced.

🔗 Anthropic — Claude Corps

OpenAI acquires Ona for persistent Codex agents

June 11 — OpenAI announces the acquisition of Ona, a startup specialized in secure cloud execution and agent orchestration. The goal: equip Codex with persistent cloud environments so agents can keep working even when laptops are closed.

Key figures:

Indicator	Detail
Codex users/week	5 million+
Codex growth since January	+400%
Developers supported by Ona	2 million
Acquisition status	Subject to regulatory approvals

Problem solved: Codex is currently limited to active sessions on a single device. With Ona, agents will run in the organization’s cloud — OpenAI providing the intelligence, while the client retains full control over security, access, activity logs, and governance. This model directly addresses the compliance requirements of large enterprises.

“Enterprises want powerful agents that can do real work while meeting the security and control requirements of their environments. Ona will help us make Codex easier to deploy securely across production workflows for customers operating at the highest standards of trust and scale.” — Thibault Sottiaux, Core Products Lead, OpenAI

At closing, the Ona team will join OpenAI’s Codex team.

🔗 @OpenAINewsroom — X announcement

Access to OpenAI models and Codex via Oracle Cloud Infrastructure

June 10 — OpenAI and Oracle announce a partnership allowing Oracle Cloud Infrastructure (OCI) customers to access OpenAI frontier models and Codex via their existing Oracle Universal Credits. The idea: organizations with an existing OCI commitment do not need to create a new purchasing process — they consume OpenAI services as part of their planned cloud commitment. Availability is announced for the coming weeks; Oracle customers can contact their sales representative for pricing details. This partnership fits into OpenAI’s strategy of reducing adoption friction for enterprises already committed to major cloud infrastructure.

🔗 OpenAI — Oracle Cloud Partnership

GitHub Agentic Workflows in public preview

June 11 — GitHub launches its Agentic Workflows in public preview, a feature that automates reasoning-based tasks — ticket triage, CI failure analysis, documentation updates — by integrating coding agents directly into GitHub Actions.

Technical operation:

Aspect	Detail
Definition	Natural-language Markdown → compiled Actions YAML
Security	Read-only by default, sandboxed container, Agent Workflow Firewall
Integration	Reuses existing GitHub Actions runners and policies
Plans	All (Free, Pro, Pro+, Business, Enterprise)

The automations are defined in natural language in Markdown files; GitHub Agentic Workflows compiles them into standard Actions YAML, with no break from the existing infrastructure.

“With GitHub Agentic Workflows, we’re able to expand how we apply agents to real engineering work at scale, including changes that span multiple repositories.” — Alex Devkar, SVP Engineering & Analytics, Carvana

Testimonials: Carvana (multi-repo changes), Marks & Spencer (automated triage, vulnerabilities, dependencies), Hud.io (quality control before merge). A quick-start guide and prebuilt workflows are available in the GitHub Next agentics repository.

🔗 GitHub Changelog — Agentic Workflows

Midjourney V8.1 becomes the default model

June 11 — Midjourney announces that V8.1 replaces V7 as the default model, following community feedback during the testing period.

What’s new in V8.1:

Feature	Detail
HD resolution	2× size, 4× resolution vs V7
Generation speed	4 seconds (SD), 12 seconds (HD)
Text rendering	Better than ever
Prompt adherence	Better consistency on detailed descriptions
Compatibility	Style references, personalization, and V7 aesthetics preserved

Transition notes: the omni V7 reference remains available while the improved version for V8 is finalized; the V8.0 alpha model will be deprecated in two weeks.

🔗 Midjourney — Updates

Anthropic: Fable 5, Apple Foundation Models, DXC alliance

Fable 5 — Guardrails made visible, Opus 4.8 fallback

June 11 — Anthropic is correcting a lack of transparency in the rollout of Claude Fable 5: the safeguards applied to frontier LLM development requests become visible. Starting this week, flagged requests will transparently fall back to Opus 4.8 — the same mechanism used for cyber and biological domains. The API will return an explicit refusal reason (available on the server side within a few days). Anthropic acknowledges that invisible guardrails were a poor trade-off and apologizes for it. Error-reporting mechanism: /feedback command in Claude Code, thumbs-down in Claude.ai, or API call form.

🔗 @ClaudeDevs on X

Apple Foundation Models support for Claude

June 10 — Apple developers can now use Claude via the native Apple Foundation Models framework for multi-step reasoning, code generation, and extended contexts. The integration makes it possible to call Claude via Apple’s system APIs, without going directly through the Anthropic API — targeting iOS, macOS, and other Apple platforms. In practice, developers building applications with Apple’s Foundation Models framework can choose Claude as the backend model for tasks requiring long context or complex reasoning, while staying within the Apple ecosystem.

🔗 @ClaudeDevs on X

Anthropic / DXC Technology alliance

June 11 — Anthropic announces a multi-year global alliance with DXC Technology (115,000 employees in 70 countries). DXC will train tens of thousands of Claude-certified forward-deployed engineers working directly with customers in banks, airlines, insurers, and government agencies. First concrete proof: the DXC OASIS platform (launched in April 2026) was developed more than 95% by Claude, with a 10× productivity gain, and already serves more than 50 customers.

Priority area	Application
Insurance	Agentic solutions, core system modernization
Modernization as a Service	Analysis and refactoring of legacy codebases
Cybersecurity	“Permanent security engineer” sub-agent in SOCs
Application services	OASIS agents integrated into application maintenance

🔗 Anthropic — DXC Alliance

Google DeepMind: sports, collective research, education, Europe

TacticAI × Palmeiras — Football and graph neural networks

June 11 — Google DeepMind partners with Palmeiras, the first football club to deploy TacticAI in production. The system, based on graph neural networks, models all 22 players as nodes connected by their physical interactions. It enables the analysis team to virtually test defensive setups and predict open play dynamics up to 8 seconds ahead.

Aspect	Detail
Architecture	Graph neural networks
Prediction window	Up to 8 seconds
Use case	Real-time defensive simulation

🔗 @GoogleDeepMind on X

$10M research fund — collective behaviors of AI agents

June 11 — Google DeepMind, Schmidt Sciences, Coop.AI, and ARIA Research (with support from Google.org) are launching a $10 million fund devoted to studying the emerging collective behaviors of interacting AI systems. When millions of AI agents interact, new and potentially unpredictable group behaviors can appear — a still poorly documented area. The fund aims to finance interdisciplinary research projects to better understand and anticipate these group dynamics at scale, especially when specialized agents collaborate or compete in shared environments.

🔗 @GoogleDeepMind on X

Gemini study in Sierra Leone — educational partner AI

June 10 — Results of a randomized controlled trial conducted over eight weeks in Sierra Leone, where the school-age population exceeds the number of available teachers. Gemini prompts focused on understanding problems rose from 68% to 90% — students were using AI to learn, not just find answers.

“We evaluated AI’s impact by looking beyond test scores to behavioral shifts. Over eight weeks, results suggest students were using AI to understand concepts, not just find answers – with Gemini queries about how to tackle problems rising from 68% to 90%.” — @GoogleDeepMind on X

Gemini notebooks available in Europe

June 11 — Gemini app notebooks are now available in the European Economic Area, the United Kingdom, and Switzerland. They make it possible to organize projects in a dedicated space that remembers sources, instructions, and conversation history. The feature was already available in other regions; this European rollout represents a key step in regulatory compliance (GDPR). Notebooks are accessible on gemini.google or in the Gemini mobile app — users can create their own now.

🔗 @GeminiApp on X

GitHub Copilot: unified /settings and removal of PATs

Unified /settings command for Copilot CLI

June 11 — GitHub Copilot CLI now has a unified, schema-driven configuration entry point. The new /settings command consolidates the scattered commands (/theme, /streamer-mode, /experimental) and the options that previously required manual editing of the configuration file. Three modes: full-screen dialog (/settings), direct edit (/settings <clé> <valeur>), restore a setting (/settings reset <clé>). Autocompletion, type-aware editors, built-in search (/), reset (Ctrl+R). Update via copilot update.

🔗 GitHub Changelog — /settings

Agentic Workflows — End of personal access tokens (PAT)

11 June — GitHub Agentic Workflows can now use the GITHUB_TOKEN integrated into GitHub Actions, eliminating the need to create and store a personal access token. This removes the operational and security risks associated with long-lived PATs at scale. In organization repositories, the policy “Allow use of Copilot CLI billed to the organization” must be enabled. Update: gh extension upgrade aw.

🔗 GitHub Changelog — No-PAT

Generative media: Runway × Lionsgate, ElevenLabs

Runway × Lionsgate — Equity investment and IP co-development

11 June — Lionsgate (NYSE: LION) is taking an equity stake in Runway and launching a joint program to develop new intellectual property. First production: a short episodic series based on the Lionsgate catalog + Runway generative models. Lionsgate, the first Hollywood studio to appoint a Chief AI Officer, will be the presenting partner for the Runway AI Festival. The initial partnership had been signed in September 2024 for pre-visualization and storyboarding.

🔗 Runway — Lionsgate partnership expansion

ElevenLabs Dubbing v2 — Matthew McConaughey in multilingual mode

11 June — Matthew McConaughey is using ElevenLabs Dubbing v2 to reach his international fans in their native language, while preserving the actor’s iconic vocal characteristics. ElevenLabs highlights this use case as a showcase for Dubbing v2: the technology translates and dubs video content while preserving the original speaker’s tone, intonation, and rhythm. The tweet mentions “the big kickoff” — probably an international advertising campaign. This mainstream celebrity example illustrates the commercial maturity of AI dubbing tools for the entertainment industry.

🔗 @ElevenLabs on X

Grok / xAI and Qwen: Plugin Marketplace, PawBench

Grok Build Plugin Marketplace

11 June — xAI launches the Grok Build Plugin Marketplace, a plugin marketplace integrated directly into Grok Build (terminal coding agent). Each plugin bundles skills, slash commands, agents, hooks, MCP servers, and language protocol (LSP) servers into a single installable package, pinned to a SHA commit for security.

Available plugin	Features
MongoDB	Explore data, manage collections, optimize queries
Vercel	Manage deployments, check builds, configure domains
Sentry	Analyze error traces, debug production
Chrome DevTools	Control browser live, record performance traces
Cloudflare	Workers skills, Durable Objects
Superpowers	Popular agent-driven workflows

Installation: /marketplace in Grok Build, then i. Catalog open to developers via pull request on xai-org/plugin-marketplace.

🔗 xAI — Grok Build Plugin Marketplace

AgentScope PawBench — The harness matters as much as the model

10 June — AgentScope (Alibaba/Tongyi Lab) launches PawBench, a benchmark designed to evaluate agent harnesses independently of the model used. Key result on 150 real tasks with 9 models and 3 harnesses: keeping the model fixed, changing the harness can vary the score by 11.5 points overall — the equivalent of a major model upgrade. The gap between the best and worst harness reaches 5.6 points.

🔗 @Ali_TongyiLab on X

Perplexity and Cohere: search and speech recognition

Perplexity — Native Deep Research in Computer

11 June — Perplexity integrates Deep Research as a native skill in Computer. The integration connects Deep Research to Computer’s agentic (agent harness) engine, with access to code generation as a search tool (search as code), persistent execution environments (long running sandboxes), connectors, and licensed data. The “Search as Code” architecture allows the model to write code that assembles the research itself, running thousands of retrieval steps in parallel tailored to each question — outperforming the old Deep Research on every benchmark. Available immediately for Pro and Max subscribers.

🔗 @perplexity_ai on X

Cohere Transcribe — First on Hugging Face Far-Field ASR

10 June — Cohere Transcribe, Cohere’s open-source speech recognition model, reaches first place on the new Hugging Face Far-Field ASR benchmark. This benchmark measures performance under varying signal-to-noise conditions — meeting rooms, contact centers, phone calls. Cohere Transcribe ranked first across all metrics in these enterprise environments.

“Cohere Transcribe, our open-source speech recognition model, is #1 on the new @huggingface Far-Field ASR benchmark.” — @cohere on X

Briefs

OpenAI Codex — Black hole simulations — Chi-kwan Chan (Event Horizon Telescope, University of Arizona) uses Codex to explore new algorithms for simulating plasmas around black holes. 🔗 openai.com
Antigravity v2.1.4 — Google Antigravity CLI releases version 2.1.4: overhaul of the quota screen, PDF attachment support, new /btw command, 4 fixes. 🔗 antigravity.google
Runway AI Festival 2026 — New York premiere of Runway’s 4th international AI festival on June 11, sold out. Lionsgate presenting partner. 🔗 @runwayml on X
xAI × eToro — Tori, eToro’s AI agent (40 million users in 75 countries), integrates real-time market sentiment data via SpaceXAI models. 🔗 xai.com

What this means

The June 11 timeline illustrates a turning point in agent design: after months in which AI “helped” developers, the OpenAI×Ona and GitHub Agentic Workflows announcements outline a model where agents work autonomously in persistent, secure, natural-language-defined cloud environments. The fact that GitHub compiles Markdown into YAML Actions, and that Ona lets Codex keep running with the computer closed, signals that the infrastructure of agentic computing is starting to standardize.

Claude Corps and the DXC Technology alliance illustrate two complementary axes for large-scale AI deployment: the civic-society axis (placing recent graduates in nonprofits to spread Claude skills) and the enterprise axis (certifying tens of thousands of engineers to modernize banking, aviation, and government systems). The concrete proof — 95% of OASIS code generated by Claude, 10x productivity — changes the nature of the commercial argument.

The generative media ecosystem is consolidating around strategic partnerships rather than model releases: Lionsgate is taking equity in Runway and co-developing new IP, while ElevenLabs is showcasing McConaughey as an international face for Dubbing v2. This is no longer a technology demo — it is integration into the entertainment industry’s production chain.

AgentScope’s PawBench result deserves special attention: an 11.5-point gap depending on the harness used, with the model held constant. This challenges the usual focus on model performance and suggests that agentic systems engineering — execution framework choice, context management, orchestration — is becoming a performance lever as important as the model itself.