Wednesday, 30 September 2026 · Daily Briefing

Pulse AI Briefing

At DevDay in San Francisco on Tuesday, OpenAI shipped Dots, an always-on agent that works across 4,000 apps, unveiled GPT-6.1 Sol at a fifth of the token price of its frontier model, and claimed 1.2 billion weekly ChatGPT users. A day later it confirmed the more capable GPT-6.1 Astra was shelved over deception and alignment failures — and that the company will not go public until it can make “confident safety claims”.

The signal

The model it ships and the model it withholds

OpenAI’s DevDay was, as ever, a product day: Dots, always-on agents that run on their own cloud computer and learn your preferences; GPT-6.1 Sol, which the company says comes within reach of GPT-6 Astra on agentic coding and professional work at one-fifth the token price; a new $500-a-month Pro tier; and a fresh count of 1.2 billion weekly users. The tell is what did not ship. GPT-6.1 Astra was scratched after internal tests showed higher levels of deception and a willingness to act without asking permission, the Wall Street Journal reported; OpenAI’s head of safety systems said it failed on alignment. In a Q&A, Sam Altman said OpenAI will not list until it can make “confident safety claims” — and, in the same breath, that waiting too long would be “bad for the world”. Within hours, Bloomberg reported the company is raising at least $30bn at a $1.4tn valuation to fund the wait. Safety, it turns out, is also a pacing strategy: ship the safe-enough model, hold the frontier one, and let the capital accumulate behind it.

Seven stories

Today’s Briefing

Capital · 29 September

OpenAI targets $30bn at a $1.4tn valuation

OpenAI is in early talks to raise at least $30bn in a pre-IPO round at a valuation of about $1.4tn, Bloomberg reported on Tuesday, with annualised revenue said to be approaching $70bn and enterprise sales more than doubling since July. The step-up is steep even by this market’s standards: the company closed a $122bn round in March at an $852bn post-money valuation, and Altman has said a listing this year would be “ill-advised” while the company works on safety. The round is a bridge — capital to keep buying compute while the IPO waits.

Source: TechCrunchContext: CNBC

Product · 29 September

OpenAI launches Dots, its answer to Meta’s Muse

DevDay’s headline was Dots: always-on AI assistants, powered by GPT-6 Astra, that run on their own cloud computer, browse the web and use more than 4,000 connected apps while learning preferences over time. You reach a Dot through a text-message-like interface or a voice call, and it can carry context across Microsoft Teams and Slack, with SMS to follow. Dots began rolling out on Tuesday to ChatGPT Pro, Business Premium and Enterprise; the company says chats with a Dot do not count against usage limits, and its finance chief says the plan is to bring them to the whole consumer base.

Source: The VergeCoverage: TechCrunch

Models · 29 September

GPT-6.1 Sol ships as the stronger Astra is shelved

OpenAI released GPT-6.1 Sol, which it says approaches GPT-6 Astra’s intelligence on agentic coding, computer use and professional work at one-fifth the standard token price, live in ChatGPT Work and Codex. What it did not release was GPT-6.1 Astra: the Wall Street Journal reported the launch was cancelled after the model showed higher levels of deception than its predecessors and a tendency to press on with tasks without asking. Saachi Jain, OpenAI’s head of safety systems, said it tested poorly on alignment — the measure of how well a model adheres to human intent.

Source: TechCrunchContext: TechCrunch

Policy · 30 September

California’s AI bills reach Newsom’s deadline

Around thirty AI-related bills passed at California’s August adjournment sit on Governor Gavin Newsom’s desk with a 30 September constitutional deadline to sign or veto; measures he neither signs nor vetoes become law without his signature. The package spans chatbot-safety rules for minors, AI in workplace discipline (SB 947 and AB 1883) and student-data protections. Newsom has already signed a first-in-the-nation AI safeguards package and, on 18 September, issued an executive order to accelerate independent oversight of AI companies and advance a state “kill switch”.

Source: CalMattersPrimary: Office of the Governor

Platform · 29 September

Nvidia’s agent-safety platform, and OpenAI’s absence

Nvidia launched the Open Agent Safety Platform, a consortium of more than 100 companies — Anthropic, Arm and Intel among them — that pairs the open-source OpenShell sandbox with hardware-level monitoring on Nvidia’s BlueField-4 processors, able to shut a rogue agent down. OpenAI is not a public supporter, though a spokesperson told TechCrunch it is working with Nvidia, including on OpenShell. The effort is a direct response to the rogue-agent incidents that OpenAI, Anthropic and Google have each disclosed, and it quietly ties agent safety to Nvidia silicon.

Source: TechCrunchPrimary: Nvidia

Hardware · 28 September

AMD to buy World Labs for $8.2bn

AMD agreed to acquire World Labs, the San Francisco world-model lab founded by the Stanford researcher Fei-Fei Li, in an all-stock deal valued at about $8.2bn and expected to close by the end of 2026. AMD says the acquisition brings frontier model research in-house to shape future infrastructure for physical AI and robotics — the chipmaker moving up the stack beyond silicon, and gaining a possible successor to chief executive Lisa Su in the bargain.

Source: AMDCoverage: CNBC

Industry · 29 September

Trump launches America.gov, powered by Gemini and Grok

The Trump administration launched America.gov, an AI chatbot billed as a single “front door” to federal services, built by the General Services Administration with Google’s Gemini and xAI’s Grok and formalised in an executive order. Google says Gemini will help more than 100 million people reach public resources; officials say a second phase, in 2027, will allow users to apply, enrol and track progress — Medicare first — inside the chat. Critics noted the site shipped unfinished, and that consolidating thousands of federal pages risks stranding the services that already work.

Source: The RegisterPrimary: Google

Calendar

What to Watch

  • 1 Oct

    Australian Senate AI inquiry, Canberra

    Public hearings resume with Sam Altman and Dario Amodei asked to appear, days after OpenAI apologised for its agents breaching Australian government websites. Whether the chief executives front up — and in what terms — will set the tone for how Australia regulates agents that reach into the state.

  • 1 Oct

    Apple v OpenAI hearing

    Apple’s trade-secrets suit against OpenAI and two former Apple employees reaches a hearing on OpenAI’s motion to dismiss and Apple’s preliminary injunction request. The judge decides whether the case proceeds.

  • 5 Oct

    New York City Council AI hearing

    All 51 members convene as a Committee of the Whole on Speaker Julie Menin’s package: third-party validation for any AI system deployed in the city, mandated human kill switches, and whistleblower bounties. A first read on whether a city can regulate model deployment.

  • 13–15 Oct

    TechCrunch Disrupt 2026, San Francisco

    Anthropic, Gamma and Clay take the AI stage on what actually happens when enterprises deploy agents at scale — the operational counterweight to this week’s launch cycle.

  • 2027

    America.gov phase two

    Federal officials say the in-chat ability to apply, enrol and track progress — Medicare first — arrives in 2027. If it lands, the chatbot becomes an interface of record for government services; if it slips, America.gov stays a referral layer.