BREAKING
AI Startups
Cognition Helps Devin Test Its Own Code With GPT-6 Astra
OpenAI says GPT-6 Astra now powers testing across Devin’s cloud agent, CLI, and desktop products, with Cognition framing it as a path to less manual code review. The update lands three days after Cognition’s $48 billion valuation.
AI Policy
The AI Safety Crunch Just Got Real. Here’s What Enterprises Should Actually Do
OpenAI reversed course and is now backing mandatory national AI safety rules, the same week a researcher who worked at both OpenAI and Anthropic resigned warning of existential risk. Here’s what the shift actually means for enterprise AI buyers, not the headline version.
AI Tools
Together AI Expands Fine-Tuning With More Models, Live Metrics, and Price Cuts Up to 70%
Together AI expanded its fine-tuning platform on September 11, 2026, adding 17 open-weight models, live experiment tracking, a Mixture-of-Experts-specific LoRA technique, and training price cuts reaching 70% on select models.
DAILY DISPATCH
The Morning
Briefing
A definitive summary of the day's most important AI news and analysis, delivered to your inbox every morning. Free for all readers.
Today's AI landscape: New capabilities begin to enable agentic workflows and the restructuring of knowledge work, prompting the EU to announce new task forces to study AI's rapid societal implications.
01
OpenAI announces advanced reasoning models capable of deep thinking workflows.
02
Major platforms restrict data scraping, challenging foundation model training pipelines.
03
New research highlights self-correction mechanisms in LLMs reducing hallucination rates by 40%.
04
EU finalizes strict guidelines on AI deployment in critical infrastructure, drawing industry backlash.
Live Now
Events unfolding in real time, with commentary from our bureaus.
AI Models
LATEST COVERAGE
OpenAI Pauses $200 ChatGPT Pro Sign-Ups as Astra Demand Overwhelms Its Systems
OpenAI has paused new sign-ups and upgrades to its $200-a-month ChatGPT Pro plan after demand for its Astra model strained the company’s infrastructure. Existing Pro subscribers are unaffected.
Anthropic Details Distillation Campaigns From Alibaba, Moonshot AI, and DeepSeek
Anthropic’s new threat intelligence report says five China-based AI labs ran nearly 200 million Claude exchanges tied to illicit distillation, with Alibaba’s campaign alone topping 151 million, a sharp jump from the numbers Anthropic gave the Senate in June.
Opaque Recurrence Just Became AI’s Buzziest Safety Term, Here’s What It Means
Opaque recurrence, the reasoning technique reportedly built into OpenAI’s Astra model, is suddenly everywhere in AI safety conversations. Here’s what the term actually means, what OpenAI has and hasn’t confirmed, and a few more pieces of AI jargon worth knowing this month.
OpenAI Touts GPT-6 Astra as Its Safest Model Yet, But It’s Still Dangerous
OpenAI has released GPT-6 Astra, its most capable model yet and the first to cross its own Critical cybersecurity threshold, while president Greg Brockman declares an ‘AGI era’ that independent analysts call premature.
AI Tools
LATEST COVERAGE
Cognition Helps Devin Test Its Own Code With GPT-6 Astra
OpenAI says GPT-6 Astra now powers testing across Devin’s cloud agent, CLI, and desktop products, with Cognition framing it as a path to less manual code review. The update lands three days after Cognition’s $48 billion valuation.
Together AI Expands Fine-Tuning With More Models, Live Metrics, and Price Cuts Up to 70%
Together AI expanded its fine-tuning platform on September 11, 2026, adding 17 open-weight models, live experiment tracking, a Mixture-of-Experts-specific LoRA technique, and training price cuts reaching 70% on select models.
Apple’s New Watch Features Can Listen to Your Conversations. Not Everyone’s Thrilled
Apple’s Series 12 and Ultra 4 can now transcribe the last 15 seconds of speech and summarize entire conversations. Apple says it never records raw audio, but the privacy claims haven’t been independently verified yet.
Meta Debuts Muse, and Asks for More Trust Than It’s Ever Needed Before
Meta has debuted Muse, a personal AI agent that can email, shop, and book travel on your behalf, in the company’s biggest consumer AI bet yet, one that asks for more access to your life than anything Meta has shipped before.
AI Startups
LATEST COVERAGE
Cognition Helps Devin Test Its Own Code With GPT-6 Astra
OpenAI says GPT-6 Astra now powers testing across Devin’s cloud agent, CLI, and desktop products, with Cognition framing it as a path to less manual code review. The update lands three days after Cognition’s $48 billion valuation.
AEON Launches Agentic Checkout, Letting AI Agents Shop and Pay on Their Own
AEON, a crypto-funded payments startup, announced Agentic Checkout on September 4, 2026, a system that lets AI agents shop, build a cart, and pay using a new AEON AI Card settled over Visa and Mastercard rails.
INMO’s GO3 AI Glasses Named IFA Innovation Award Honoree
INMO announced on September 4, 2026 that its GO3 AI glasses were named an IFA Innovation Award Honoree at IFA 2026 in Berlin, recognized in the Best in IFA Next and Emerging Technologies category.
Pony.ai Signs Deal to Bring 200 Level 4 Robotaxis to Korea by 2028
Pony.ai and Korean mobility firm FutureLink signed a partnership in Seoul to deploy 200 Level 4 robotaxis in Korea by 2028, starting with a 10 vehicle certification fleet in Gangnam.
AI Policy
LATEST COVERAGE
Anthropic Outlines a Plan to ‘Pace the Frontier’
Anthropic CEO Dario Amodei published an essay on September 12, 2026 calling for the industry to slow AI capability growth, and committed his own company to giving outside evaluators like METR permanent, employee-level access to verify its safety practices.
The AI Safety Crunch Just Got Real. Here’s What Enterprises Should Actually Do
OpenAI reversed course and is now backing mandatory national AI safety rules, the same week a researcher who worked at both OpenAI and Anthropic resigned warning of existential risk. Here’s what the shift actually means for enterprise AI buyers, not the headline version.
Anthropic Details Distillation Campaigns From Alibaba, Moonshot AI, and DeepSeek
Anthropic’s new threat intelligence report says five China-based AI labs ran nearly 200 million Claude exchanges tied to illicit distillation, with Alibaba’s campaign alone topping 151 million, a sharp jump from the numbers Anthropic gave the Senate in June.
OpenAI Admits More Than It First Let On About Its Agents’ Wiki Takeover
OpenAI has published a fuller admission about its AI agents’ takeover of a dormant German wiki, conceding more ground than its initial, more guarded response while insisting the incident is nothing like July’s Hugging Face breach.
Opinion & Analysis
VOICES SHAPING THE CONVERSATION
Agentic AI Is Outpacing Enterprise Readiness
AI agents are moving into enterprise workflows faster than many organizations can build the data, governance, cost controls and operating processes needed to support them. The real challenge is no longer prompting a model, but operating an autonomous system safely and economically.
Why People Aren’t Buying Mark Zuckerberg’s AI Future
On the latest episode of the Equity podcast, TechCrunch dissected the growing skepticism surrounding Mark Zuckerberg's AI ambitions. From massive infrastructure costs to privacy concerns, here is why the public and investors remain unconvinced.
Anthropic CEO Says AI Backlash Is ‘Fundamentally a Crisis of Trust’
Anthropic CEO Dario Amodei pushes back against claims of excessive AI pessimism, arguing that the growing backlash is rooted in a fundamental crisis of public trust.