AI Product Updates Daily — June 13, 2026

June 12 was Anthropic's biggest single day in months — the company launched a 1,000-person national fellowship, published its most substantive governance essay to date, and reversed a widely criticized safety policy all in the same window. Meanwhile, OpenAI's ChatGPT crossed 1 billion monthly app users even as the company is reportedly weighing a major API price cut to slow Anthropic's momentum. xAI shipped its plugin marketplace for Grok Build, DeepMind and partners committed $10M to multi-agent safety research, and enterprise deals kept landing: TCS is putting Claude in front of 50,000 employees.

Here is everything that changed on June 12, 2026.

Anthropic launches Claude Corps and makes Fable 5 guardrails visible

The biggest item of the day is Claude Corps, Anthropic's new national fellowship program. Starting this week, Anthropic will recruit 1,000 early-career fellows, pay each one a full-time $85,000 salary, match them with nonprofits across America, and have them spend a year deploying Claude on-site. The company is committing an initial $150 million to the program. 1

The program runs as a three-way partnership. Anthropic funds and shapes strategy. CodePath — described as the largest provider of collegiate computer science education in the US — acts as the official employer of record and runs training. Social Finance, a nonprofit registered investment advisor, handles measurement and will build a longer-term funding vehicle if the program scales. At least 400 nonprofits will host fellows in the first cohort, including food banks, veteran service organizations, and college access nonprofits.

Fellows get five hours of training per week, a Claude token budget, a CodePath mentor, and direct office hours with Anthropic staff. On the same day, CEO Dario Amodei published what the LoreAI briefing described as "the most detailed AI governance blueprint from any lab CEO" — a full essay laying out concrete mechanisms for lab self-regulation, compute governance, and international coordination before governments impose their own rules. 2

Separately, Anthropic walked back the opaque refusal behavior in Claude Fable 5 that had frustrated developers for weeks. When Fable 5 was released, it silently routed certain requests — including prompts related to distillation, cybersecurity, and biology — to the older Claude Opus 4.8 without telling users. After sustained pushback from the developer community, Anthropic announced that flagged requests will now visibly fall back to Opus 4.8 every time it happens; the same visibility standard already applied to cyber and bio guardrails. The underlying guardrails themselves remain in place. 3

正在加载内容卡片…

TCS partnership: Announced June 11 but still circulating widely today, Tata Consultancy Services signed a Global Premier Partnership with Anthropic. TCS will equip 50,000 employees across engineering, finance, legal, marketing, and sales with Claude through enterprise-wide licensing, and will stand up a dedicated business unit for Claude-based customer solutions. 4 A separate Anthropic–DXC alliance announced this week targets regulated sectors — banks, airlines, and government — where AI adoption has stalled over compliance concerns. 5

OpenAI: ChatGPT hits 1 billion monthly users, API price cut under consideration

ChatGPT reached 1 billion monthly active app users in May 2026, according to Sensor Tower estimates — the fastest any app has ever reached that milestone, surpassing the previous record set by Google Maps. 6

The growth picture is more complicated under the surface. ChatGPT's monthly active user base grew 62% year-over-year, but rivals grew faster off a smaller base: Claude grew 640% year-on-year and Meta AI grew 973%, per the same Sensor Tower data. OpenAI held roughly six times the web traffic and mobile sessions of its next largest AI platform as of February, when it reported 900 million weekly active users across web and mobile.

The billion-user number arrives as OpenAI is actively weighing a major token-price reduction. According to a Wall Street Journal report, the company is evaluating cuts to API token pricing — the per-million-token rates developers pay — with people familiar with the matter saying Anthropic may follow with its own reductions. 7 Sam Altman has previously stated publicly: "I think we'll have a lot of ways we can help people get more value for less spend." The cuts remain unconfirmed and the extent is unknown. Consumer subscription pricing is not reported to be part of the consideration.

The timing is deliberate. Anthropic surpassed OpenAI's valuation for the first time, reaching $965 billion at its recent IPO filing, and has been winning enterprise deals. OpenAI is also working to expand Codex and is planning a ChatGPT "Superapp" that would combine image generation, Codex, and other services in one product.

ChatGPT model picker redesign: OpenAI also redesigned its model selection interface in the app this week after user complaints about the previous layout. 8

xAI ships Grok Build Plugin Marketplace with MongoDB, Vercel, Sentry, and more

xAI launched the Grok Build Plugin Marketplace on June 11, adding a discovery and distribution layer to its terminal coding agent. 9

A plugin bundles skills, slash commands, agents, lifecycle hooks, MCP servers, and LSP servers into a single installable package. Before this, developers had to wire up each integration separately. The launch catalog ships six plugins from named partners:

Plugin	What it does
MongoDB	Explore data, manage collections, optimize queries
Vercel	Manage deployments, check build status, configure domains
Sentry	Analyze stack traces, debug production errors
Chrome DevTools	Control a live browser, record performance traces, inspect network requests
Cloudflare	Workers, Durable Objects, and related skills
Superpowers	Popular agent-driven workflow bundles

正在加载内容卡片…

The marketplace is accessible inside Grok Build via /marketplace and via grok plugin install. Every remote plugin pins a 40-character commit SHA, which Grok Build re-verifies after cloning as a supply-chain guard. Third-party plugins are listed as community-contributed: xAI verifies the SHA pin but explicitly does not author, control, or verify plugin behavior. The catalog is open to external contributions via pull request against xai-org/plugin-marketplace on GitHub.

Access still requires a paid SuperGrok or X Premium Plus subscription.

DeepMind and partners commit $10M to multi-agent AI safety research

Google DeepMind, Schmidt Sciences, the Cooperative AI Foundation, ARIA, and Google.org announced a $10 million funding call for researchers worldwide working on multi-agent AI safety. 10

The call targets a specific and largely unstudied risk: when large numbers of AI agents built by different organizations interact at scale, emergent collective behaviors can arise that don't exist in any single model and that current safety evaluations — which test models in isolation — cannot predict or detect. DeepMind's announcement describes potential risks from these emergent dynamics including "unpredictable flurries of economic activity" and new security challenges.

The four funded research areas are: building sandboxes and testbeds for realistic multi-agent evaluation; studying the science of agent networks (how collective behaviors emerge and how networks fail); stress-testing agent identity and reputation infrastructure; and developing oversight methods for deployed agent populations. Proposal deadline is August 8, 2026, with awards announced in Autumn 2026.

A row of translucent colored glass cubes with white light passing through them, illustrating multi-agent interaction dynamics. — Multi-agent AI safety research — image from Google DeepMind 10

Gemini on Google TV: Google also rolled out voice-controlled settings for Google TV via Gemini. Users can adjust brightness, contrast, picture mode, volume, sound mode, and bass by describing what they want, and Gemini will apply the change. The rollout is live for US users on select TCL models (QM9K, X11L, QM9L, QM8L, RM9L), with broader availability planned. 3

Quick updates

Claude Code v2.1.173 — released June 12, fixes Fable 5 model name normalization and a Windows sandbox warning. 11
Vercel AI SDK critical patch — the SDK was executing tool calls from client message history without re-validating input schemas or approval status. Version ai@6.0.202 patches the validation bypass. Any production workflow using tool approvals should update immediately. 12
SCAIL-2 — an open-source image-to-video model published on HuggingFace by zai-org, joining the expanding field of open-weight video generation. 13
YouTube DMs — YouTube is rolling out in-app video sharing and messaging to US users and globally. The feature requires age verification (18+) and an invite link to start a conversation. Messages are not end-to-end encrypted. 14
DeepMind TacticAI + Palmeiras — Brazilian football club Palmeiras is reported as the first football club building on DeepMind's TacticAI system, which simulates play dynamics up to eight seconds ahead for real-time tactical prediction.
Prometheus valued at $41B — Jeff Bezos's AI startup (industrial AI, not consumer chatbots) is reportedly valued at $41 billion in recent funding discussions. 15