A short daily roundup of relevant AI news from roughly the last 24–48 hours. Stories already covered in yesterday’s roundup (Gemini 3.8 Live, OpenAI–Anthropic–DeepMind safety talks, Musk’s peer-review proposal, TotalEnergies×Mistral) are not repeated; below are the new announcements and the continuation of the safety debate.
Mistral × Mozilla: Firefox Smart Window runs on Mistral models. On September 16 Mistral and Mozilla officially announced the partnership: Firefox’s AI browser assistant (Smart Window beta) will use Mistral models — specifically Mistral Small 4, according to Mozilla. Target markets: the US, Canada, and France (with French language support); the UK and Germany are expected later. Both sides emphasize model choice, multilingual support, and privacy: conversations are not saved to Mozilla’s servers by default, and partners (including Mistral) commit to zero data retention. Source: Mistral, Mozilla Blog.
Cohere + Aleph Alpha: definitive merger agreement — “transatlantic sovereign AI.” Canada’s Cohere and Germany’s Aleph Alpha signed a definitive business combination agreement on September 16 (continuing the April plan). The company continues under the Cohere brand, with dual headquarters in Toronto and Berlin; Aleph Alpha’s Heidelberg office remains a research hub. According to Reuters the combined company is valued at about $20 billion; the Schwarz Group is investing €500 million, and via the STACKIT cloud a compute investment of as much as €11–13 billion is also mentioned. The deal awaits regulatory approval, with closing expected later in the year. Source: Cohere, Reuters.
xAI: Grok Build gets memory. Official announcement on September 16: Grok Build now preserves conventions, decisions, and project facts across projects. After a turn, markdown notes are created in the background;
/dreamorganizes them into themes, and/memoryshows them in a browser. A new session is required (/newor a freshgrok); in-conversation instructions override the notes. Source: xAI / SpaceXAI News.OpenAI: ChatGPT Ads — Sponsored Agents, Ads Manager AI, HubSpot and Shopify. On September 16 OpenAI introduced new AI experiences for the ChatGPT advertising platform: Sponsored Agents let the user start a separate, clearly labeled conversation with the business’s sponsored agent after an ad (testing in the US with selected advertisers). The ChatGPT Work + Ads Manager plugin offers natural-language campaign creation, analysis, and creative suggestions; AI body copy customization and translation are coming. First CRM/ecommerce integrations: HubSpot and Shopify (Shopify on international markets with ChatGPT Ads availability, from September 23). Source: OpenAI.
Clarification / continuation of our earlier post: Latest AI news — September 15, 2026 — Mark Zuckerberg (Meta) on Tuesday rejected coordinated industry slowdown in an X post: in his view every lab has the incentive and responsibility to proceed at its own safe pace; liability and competition are strong enough. He does support independent evaluators (“industry best practice”), and writes that Meta Superintelligence Labs already uses them in several areas; he cites the delayed Muse AI release as an example. Most compute goes to user products, not recursive self-improvement. This is Meta’s opposite stance to the Amodei/Altman pacing line and to the OpenAI–Anthropic–DeepMind safety talks described yesterday. Source: Reuters, CNBC.
EU: Ribera open to treating AI-safety coordination under antitrust — but no formal request yet. EU competition commissioner Teresa Ribera said on Wednesday she would examine it if frontier labs asked for antitrust room for safety cooperation; so far “we have not received a formal request.” In the State of the Union speech, von der Leyen talked about slowing self-recursive models and promised talks with the labs. Reception in Washington is cooler: FTC chair Andrew Ferguson called the antitrust-exemption + regulatory-lobby combination “deeply suspicious.” Source: POLITICO Europe.