One-Line Summary

At the G7 on Jun 17, Big Tech leaders pressed for a U.S.-led AI rulemaking bloc as a new startup raised $27M to make AI outputs verifiable.

Big Tech

Anthropic and DeepMind urge U.S.-led AI coalition at the G7

Two top AI lab CEOs used the G7 summit to ask governments to form a U.S.-led group to set shared AI rules and standards. At a closed-door lunch in Évian-les-Bains on Jun 17, 2026, Anthropic’s Dario Amodei and Google DeepMind’s Demis Hassabis made the pitch to heads of state and tech leaders, with OpenAI’s Sam Altman also in the room, aiming to coordinate protections against emerging AI risks. ¹

According to people familiar with the meeting cited by CNBC, Amodei proposed structured access to frontier models and chip and component trade that excludes China, alongside cooperation on cyber, bioterrorism, and intelligence risks. Canadian Prime Minister Mark Carney agreed that the U.S. could lead an AI coalition, according to the report. ¹

The push comes amid U.S. export controls on Anthropic’s newest models: the company disabled access to Fable 5 and Mythos 5 on Jun 12, 2026 and remains in negotiations with the Trump administration. Altman called for an international forum to set globally accepted testing standards and provide impartial capability and risk analysis; OpenAI previously announced on May 7, 2026 that GPT-5.5 Cyber would roll out in limited preview to vetted cybersecurity teams. ¹

Analysts told CNBC the CEOs’ seats at the G7 signal rising geopolitical influence for private AI labs, with other attendees including leaders from Mistral, Cohere, Synthesia, and Black Forest Labs. The Élysée Palace said topics include frontier risks, infrastructure, sovereignty, and child protection, and commentators noted companies expect a package of voluntary commitments on youth safety and frontier risks. ²

Industry & Biz

Pramaana Labs raises $27M to apply formal verification to AI

Pramaana Labs is building a layer that checks AI answers against codified rules, combining a large language model with a deterministic verification system. The startup raised $27 million in seed funding led by Khosla Ventures, targeting high-stakes uses like law, drug discovery, and tax, and drawing on the LEAN programming language used to verify mathematical proofs. ³

For each domain, Pramaana plans to formalize the rules with experts — including former IRS commissioner Danny Werfel for tax — and cites France’s CATALA project as precedent for turning policy into executable code. The approach aims to curb hallucinations and make AI decisions auditable in regulated workflows. ³

What This Means for You

A U.S.-led AI coordination push at the G7 puts testing, safety, and access control on center stage for enterprises using AI. For teams handling security-sensitive or regulated tasks, this elevates the need to document how models are tested and who gets access to the most capable systems. ¹

Export controls already affect product continuity: Anthropic disabling access to Fable 5 and Mythos 5 on Jun 12 shows vendors may limit features under government orders. If your tools rely on frontier capabilities, plan for alternates or downgraded modes. ¹

Reliability investments are moving toward verifiable outputs. Pramaana’s funding underscores a pragmatic playbook for business use: encode key rules and validate AI outputs against them to reduce costly errors in areas like tax, claims, or compliance reviews. ³

The testing conversation is getting institutional backing. OpenAI’s call for an international forum to set shared testing standards suggests hiring and upskilling around evaluation, red-teaming, and audit trails will help teams align with emerging norms. ¹

Action Items

1. Run a quick AI risk map: List where your workflows use AI for decisions in security, finance, legal, or compliance, and note which features rely on “frontier” capabilities.
2. Ask vendors about controls and fallbacks: Request a short note on any export-control exposure of their models and what continuity plan (alternative models or modes) they provide.
3. Add a simple rules check to one workflow: Write 10 business rules (e.g., tax or policy criteria) and have AI produce structured fields you can verify against those rules before acting.
4. Create a 20-case evaluation set: Save real prompts and expected answers for your top use case, and re-run them weekly to track accuracy and spot regressions.