AI News September 17, 2026: EU Chief Von der Leyen Endorses 'Pace the Frontier', Anthropic Fuses Claude Chat and Cowork, Cohere–Aleph Alpha $20B Merger Signs, Bengio's LawZero Lands $300M, and ChatGPT Shows Its First Sponsored Agents

One day after the three biggest US labs confirmed they are coordinating on safety, the pacing debate jumped the Atlantic — and the product and money news refused to wait for it.

1. Von der Leyen Endorses “Pace the Frontier,” Calls Brussels Summit

European Commission President Ursula von der Leyen used her State of the Union address in Strasbourg to frame AI as one of the two “tipping points of our time” alongside climate change — and to side, explicitly, with the labs urging restraint.

Quoting the industry directly — “CEOs of the most advanced companies tell us that it is time to slow down… To pace the frontier” — she endorsed the call that Dario Amodei made in his September 12 essay and that OpenAI and Microsoft have since backed. She announced she will invite the main frontier AI labs to Brussels to discuss risk-mitigation efforts, giving the EU a seat in a conversation that until now ran through Washington.

2. Anthropic Merges Claude Chat and Cowork Into One Interface

Anthropic combined Claude chat and Cowork into a single interface that automatically routes requests across chat, Artifacts, and Claude Design — no more tab switching. The update adds presentation creation with PDF and PowerPoint export plus collaborative document editing, rolling out first to Pro and Max subscribers on web, desktop, and mobile before reaching free and team tiers.

The merge turns Claude from a chatbot with attached tools into a workspace competitor, and the timing is pointed: it lands the same week OpenAI confirmed staff friction over pacing and as Anthropic’s compute commitments hit $517 billion. Slides and docs are where enterprise budgets actually live.

3. Cohere and Aleph Alpha Sign Definitive $20B Merger

Cohere and Aleph Alpha signed their definitive combination agreement, formalizing the April framework into a dual-headquartered Berlin–Toronto entity with more than 1,000 staff. Aleph Alpha co-CEO Ilhan Scheer becomes Cohere’s COO, and Schwarz Group is investing $600M to lead a Series E reportedly valuing the combined firm at roughly $20B.

The deal creates the largest explicitly sovereignty-focused frontier lab outside the US and China — data-resident, EU-governed, and now funded like a top-tier contender. With von der Leyen calling for European AI strength the same day, the merged company becomes the continent’s flagship bet.

4. Canada and Germany Put $300M Into Bengio’s LawZero

Canada and Germany each committed up to $150M in grants to LawZero, Yoshua Bengio’s Montreal-based non-profit, unveiled at Montreal’s All In AI conference. The funding underwrites hiring and compute for “Scientist AI” — a monitoring-guardrail system Bengio argues can flag misaligned behavior without relying on reinforcement learning.

Governments directly funding an independent safety organization, rather than regulating labs into funding one, is a new model — and it lands while the FRONTIER Act’s independent-verifier provisions move through Congress.

5. OpenAI Pilots Sponsored Agents in ChatGPT

OpenAI began testing Sponsored Agents inside ChatGPT: clearly labeled, advertiser-run conversational agents that answer follow-ups and hand off to a business’s site. Wayfair and Angi are among the first participants. The company simultaneously opened ChatGPT Ads integrations with HubSpot for lead management and a Shopify App Store connector for US merchants.

This is the first concrete look at advertising in an agentic interface — not banner ads, but sponsored counterparties inside the conversation, arriving as ChatGPT crosses 900M users.

Also this week

• Suleyman vs. model welfare: Microsoft AI’s CEO published an essay arguing Anthropic’s consciousness language in Claude’s constitution is circular reasoning that could make future AI “harder to turn off.”
• Anthropic goes Down Under: a A$32B, 2.16GW Queensland data-center lease for Claude inference from 2027.
• Biren’s next billion: the Chinese GPU maker weighs its third $1B share placement since its January Hong Kong IPO.
• Benchmark warning: the ImpossibleRubrics paper shows LLM-generated rubrics get gamed 8–26% as RL rewards — a sharp caveat for automated evals.

The takeaway: regulation, product, and capital are now moving in the same direction at the same speed — the frontier is being paced, funded, and monetized all at once.