The three biggest AI labs are heading to the G7

Good morning 👋 The heads of OpenAI, Anthropic, and Google DeepMind are about to sit at a G7 table with the people who'd regulate them. First time all three have shown up at once.

In today's issue:

• 🔭 All three frontier labs are heading to the G7 summit
• 🧠 Models & releases: Claude Mythos 5 goes GA, Meta's Muse Spark
• 💰 Business & funding: OpenAI files to go public, Anthropic takes the valuation crown
• 🛠️ Tools & how-to: a 20-minute dependency triage you can run this week
• 🔗 Quick links: Novo Nordisk x OpenAI, seven new Microsoft models, Glasswing at 200

🔭 THE ONE THING

🏛️ The three biggest AI labs are heading to the G7

What happened: France's presidency published the guest list for next week's G7 at Évian (June 15 to 17), and it includes Sam Altman, Dario Amodei, and Demis Hassabis (The Next Web). Bloomberg reported it first. It's the first G7 with the heads of OpenAI, Anthropic, and Google DeepMind in the room together, and they won't be alone: Mistral's Arthur Mensch, Cohere's Aidan Gomez, Meta's Alex Wang, and Salesforce's Marc Benioff are on the list too. A Wednesday working lunch puts them across from heads of state, with AI infrastructure, networks, and regulation on the agenda.

Why it matters: For two years the labs lobbied from the hallway. Now they're invited guests. That cuts both ways. They get access to the room where the rules get drafted, and governments get to treat a handful of private companies as something close to state actors on infrastructure and security. Don't expect binding commitments this round. Nobody signs a treaty over a working lunch, and the likely output is another batch of voluntary pledges. The signal is the seating chart, not the communiqué.

The takeaway: If you ship on these models, the terms you operate under are increasingly set in rooms like this one, well upstream of any product doc.

🧠 MODELS & RELEASES

• 🛡️ Anthropic made Claude Mythos 5 generally available, the first Mythos-class model it has shipped with full safeguards rather than as a gated preview (Anthropic). It's the same model line behind Project Glasswing, the security effort whose partners have now flagged more than 10,000 high and critical vulnerabilities in real infrastructure.
• 🧪 Meta unveiled Muse Spark, the first flagship model out of Alexandr Wang's new Superintelligence Labs, and paired it with a 2026 capex plan of $115 to $135B, roughly double last year (CNBC). The strategy reads as: spend your way back into the race.

💰 BUSINESS & FUNDING

• OpenAI filed a confidential S-1 on June 8, with Goldman Sachs and Morgan Stanley leading and a target valuation north of $1 trillion (The Statesman). Run-rate revenue is around $25B. The company was careful to say timing isn't set and a listing "may be a while." The asterisk worth keeping in view: internal projections still point to roughly a $14B loss this year.
• Anthropic quietly became the most valuable private AI company after a $65B Series H put it near $965B in late May, the first time anyone has edged past OpenAI on the private market (context).

🛠️ TOOLS & HOW-TO

Triage your own dependencies before someone else does

With Mythos-class models now finding zero-days at scale, the bar for "secure enough" just moved. A pass you can finish in 20 minutes:

1. List your top direct dependencies (npm ls --depth=0, or the equivalent for your stack).

2. Paste that list plus your lockfile into a frontier model and ask for the riskiest packages by transitive blast radius, not raw CVE count.

3. Pin or replace the single worst offender today. File issues for the rest.

Prompt: Here is my dependency list and lockfile. Rank the 5 packages with the largest transitive attack surface (dependency depth, maintainer count, time since last release). For each, give one concrete mitigation I can ship this week.

5 trending right now (open-source agents)

• 🦅 OpenClaw: local-first personal agent you drive from your chat apps; past 145k stars (GitHub).
• 🌐 browser-use: hand an agent a real browser without the glue code (GitHub).
• 👥 CrewAI: orchestrate role-playing crews of agents on one task (GitHub).
• 🔧 Dify: self-hostable platform for shipping LLM apps fast (GitHub).
• 🧱 Langflow: drag-and-drop builder for agent pipelines (GitHub).

🔗 QUICK LINKS

• Novo Nordisk signed OpenAI across drug discovery, clinical trials, and manufacturing, with full rollout targeted by year-end. Big-pharma AI is getting past the pilot stage (link).
• Microsoft shipped seven in-house MAI models (reasoning, coding, image, voice, transcription), claiming up to 10x cost efficiency on some tasks (link).
• Anthropic widened Project Glasswing to roughly 200 organizations across 15-plus countries, leaning into power, water, and healthcare infrastructure (link).

See you tomorrow.

Pradeep Perugu