← Dailies

Daily · 2026-09-17 00:00 UTC

Anthropic merges Claude Chat and Cowork; DeepMind opens an AGI institute — 2026-09-17 00:00

Anthropic folds Claude Chat, Cowork and Design into one product, OpenAI tests sponsored agents inside ChatGPT ads, Google DeepMind launches the DeepMind Institute, and AWS confirms permanent data loss in its Bahrain region.

Source ↗

Ten stories from the last 24 hours. The through-line: everyone is racing to own the layer where agents actually live — one product instead of five, ads that talk back, a smart home wired for agents, and a reminder that infrastructure promises do not survive a war.

1. Anthropic merges Claude Chat, Cowork and Design into "one Claude" — Anthropic has begun folding Claude Chat, Cowork and Design into a single product so users no longer have to pick a tool per task; Claude now decides whether a request needs a quick answer or a longer workflow. Rolling out on Pro and Max over the next few weeks, then Team and Free, with Enterprise following at least 30 days after admins are notified. Claude Docs and Claude Slides are in beta on paid plans. PYMNTS · Sep 16
2. OpenAI tests business-sponsored AI agents inside ChatGPT ads — OpenAI is testing "Sponsored Agents": after seeing a relevant ad in ChatGPT, a user can start a clearly labeled conversation with a business-sponsored agent, ask follow-up questions and follow a link to the business's site. The conversations are separate from the original ChatGPT thread and from ChatGPT's own answers, and the test runs with select US advertisers. OpenAI is also adding prompt-based ad creation in ChatGPT Work, new creative tools in Ads Manager, and integrations with HubSpot and Shopify. PYMNTS · Sep 16
3. Google DeepMind launches the DeepMind Institute to study AGI — The new DeepMind Institute (DMI) is an interdisciplinary platform for research and debate on artificial general intelligence, drawing on researchers from DeepMind, Google and the global scientific community, plus the arts, humanities and policy. Directors are Shane Legg, James Manyika and Nobel laureate Demis Hassabis; the stated focus is safety, governance and risks including cyberattacks and loss of control. Legg has said a precursor could arrive by 2028. The Decoder · Sep 16
4. Apple reportedly building an enterprise AI inference server on M8 Ultra chips — According to The Information, Apple is developing an enterprise server in two versions, with either two or four M8 Ultra chips, aimed at AI developers, businesses and governments and built for inference on already-trained models. Apple is considering Nvidia's NVLink Fusion to connect the chips. Launch is not expected before 2029, and the project could still be scrapped. The Decoder · Sep 16
5. Anthropic lands deal for a $31.9bn data centre in western Queensland — Anthropic has done a deal to lease the site of its first Australian data centre, a $31.9bn project that planning documents describe as the largest of its kind in Australia. Queensland premier David Crisafulli announced the investment in parliament on Wednesday. The Western Downs digital park is to be built by Singapore-based Zerra DC near Dalby and will connect to the Braemar power station; the first stage is planned to open next year, and the project has yet to be approved by the local council. The Guardian · Sep 16
6. Snap ships its first consumer AR glasses at $2,200 — Snap's Specs are the company's first consumer augmented reality glasses after roughly half a decade of work. The Verge's hands-on describes a demo of playing virtual dominoes with a Snap employee wearing the same pair: you select a virtual piece with a pinch and move your hand to drop it. The glasses are chunky and expensive, and the reviewer calls them the boldest hardware bet of the week. The Verge · Sep 16
7. Fountain 0's AI-generated feature film lands with a thud — Odysseus: The Fall is a 2.5-hour film made entirely with AI, written and directed by Fountain 0 cofounder Ash Koosha, who also lent his likeness to the title character and wrote all the music. It is billed as "the first completely AI generated film produced at the level of a Hollywood blockbuster." The Verge's review is blunt: it is nowhere near that level, and the runtime is a big part of the problem. The studio's only previous work, Dreams of Violets, screened at Tribeca earlier this year. The Verge · Sep 16
8. Google Home opens an MCP server to AI agents — Google rolled out early access to a Model Context Protocol server for its Google Home ecosystem, letting any MCP-capable agent — Claude, ChatGPT and others — control connected devices, review camera summaries, read event history and build custom smart home dashboards using natural language. Setup requires a Google Cloud project configured for Home MCP. Access starts with US subscribers to Google Home Premium Advanced, the $20-per-month tier, rolling out over the coming weeks. TechCrunch · Sep 16
9. Google patches a Pixel modem zero-day exploited in targeted attacks — Google says a bug in Pixel smartphone modems, tracked as CVE-2026-58704, was exploited in limited and targeted attacks and has now been patched. Exploiting it could let an attacker escape the modem's sandbox into the wider phone's data — privilege escalation — and it could be triggered silently, with no interaction from the owner. Google did not say who was exploiting it. TechCrunch · Sep 16
10. AWS can't restore data lost in its Bahrain region after Iranian strikes — In an update to its Health Dashboard, AWS confirmed that resources and data hosted exclusively in its Bahrain region (me-south-1) remain inaccessible and cannot be restored: "The damage to our infrastructure spanned multiple Availability Zones and exceeded what our regional and multi-AZ services are designed to withstand." One availability zone in the UAE (mec1-az2) is also unrecoverable. The Register · Sep 16

🧪 Try This Next

• 🔧 Read OpenAI's new framework for reporting model misalignment — six incidents disclosed since March, and it is the template regulators are likely to copy
• 🎥 Watch the Emergence research on agent dialects — if you deploy agents, monitoring is the bottleneck
• 📝 Note CVE-2026-58704 — the Pixel modem privilege-escalation pattern is a useful reminder to track zero-click mobile bugs if you manage fleets

Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.