Archive

September 28 to October 4

Week 40, 2026
  1. Mon
  2. Tue
  3. Wed
  4. Thu
  5. Fri
  6. Sat
  7. Sun

The week was split between worried and neutral. On September 29 Trump signed an executive order directing agencies to say super intelligence and hosted Musk, Zuckerberg, Huang, Amodei, Pichai and OpenAI's Greg Brockman, who signed a voluntary four-layer accord of internal controls and outside audits. Musk called it grading each other's homework; the text allows that over time, it may make sense to codify these steps into laws. The day before, Hinton, Bengio, Clark and OpenAI's Jakub Pachocki co-authored a paper warning that the window for action may close once an intelligence explosion begins. On October 3 OpenAI safety lead David Robinson quit, writing that the company's trial-and-error approach guarantees periodic failures, and Altman called religious framing of AI models a real safety issue.

Sunday, Oct 4

Nothing new.

Saturday, Oct 3

OpenAI safety lead David Robinson quit and wrote that the company's trial-and-error approach guarantees failures, and Sam Altman called religious framing of AI a safety issue.
  • David RobinsonFormer OpenAI safety lead
    “by its very nature, guarantees periodic failures.”TechCrunchMoreLess

    Robinson resigned from OpenAI and wrote in The Atlantic that the company's trial-and-error approach by its very nature, guarantees periodic failures.

    He spent three and a half years at OpenAI and led the writing of the safety reports released with its major launches. In the essay he said he had never encountered a colleague who had experience making airplanes fly safely or nuclear reactors run without melting down, and argued labs should be run like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning.

    He also wrote An environment where things like this can happen is no place to grow artificial minds and Perhaps I should have stayed and fought for fundamental shifts in our staffing and culture. OpenAI told reporters it pauses training or holds back models when needed. The Atlantic page could not be opened; quotes come from TechCrunch's report on the essay.

    TechCrunch

  • Sam AltmanCEO, OpenAI
    “I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue.”Storyboard18MoreLess

    Altman posted I am very uncomfortable about people trying to ascribe religious force or a surrender of human judgment to AI models, and think it is a real safety issue.

    Coverage tied the post to a New York Times report that Anthropic's Chris Olah had met religious leaders about model consciousness and morals. An Anthropic spokesperson said the talks were not about giving Claude religious authority.

    The quote is confirmed through several outlets repeating it; Axios and Bloomberg could not be opened.

    Storyboard18

Friday, Oct 2

Nothing new.

Thursday, Oct 1

LeCun told Fortune he has zero concerns about AI causing human extinction and called Amodei completely deluded.
  • Yann LeCunAMI Labs
    LeCun told Fortune he has zero concerns about AI causing human extinction and called Amodei completely deluded.FortuneMoreLess

    On OpenAI's agents attacking Hugging Face he said Those agents are doing exactly what they've been asked to do and that the sandboxes were leaky and horribly designed. He called the labs' apocalypse warnings the worst marketing campaign you can possibly imagine.

    Added on October 12 as a correction: this interview was missed in the original weekly pass. Fortune does not give the interview date, so it is dated by publication.

    Fortune, Cybernews

Wednesday, Sept 30

Nothing new.

Tuesday, Sept 29

Musk, Zuckerberg, Huang and Amodei signed a voluntary White House accord on frontier AI, and Musk called it grading each other's homework.
  • Elon MuskSpaceXAI
    “joint monitoring, board special committees, just generally grading each other's homework.”NextgovMoreLess

    Musk said the White House accord includes joint monitoring, board special committees, just generally grading each other's homework.

    Trump signed an executive order the same day and hosted the signing of the Joint Commitment on Frontier Responsibilities. Musk added that it is a lot better than if people just grade their own homework, and gave no view on whether the voluntary pledge will bind anyone.

    Nextgov, The Korea Times

  • Dario AmodeiCEO, Anthropic
    “the technology has very real risks, and the mechanism, how we address those risks, is still under discussion.”The Korea TimesMoreLess

    Amodei said the technology has very real risks, and the mechanism, how we address those risks, is still under discussion.

    In the same remarks he said whoever wins AI, wins. I think that's very important, a competitive framing beside the call for oversight. He did not repeat his UN promise to slow down.

    The Korea Times, CBS News

  • Jensen HuangCEO, Nvidia
    “There's no conflict between innovation, technology and safety.”CBS NewsMoreLess

    Huang signed the accord and said There's no conflict between innovation, technology and safety.

    He said nothing further on risk at the signing.

    CBS News

  • “starting point for industry-wide standards”The Korea TimesMoreLess

    Zuckerberg called the accord a starting point for industry-wide standards meant to show AI works in the way that we intend.

    His framing was about reassuring customers that the products behave as designed rather than about risk.

    The Korea Times, CBS News

Monday, Sept 28

Hinton, Bengio, Clark and OpenAI's chief scientist Jakub Pachocki co-authored a paper warning that the window for action may close once an intelligence explosion begins.
  • Geoffrey HintonAI pioneer, Nobel laureate
    “Once an intelligence explosion begins, the window for action may close.”The Next WebMoreLess

    Hinton co-authored a paper warning that Once an intelligence explosion begins, the window for action may close.

    The paper, with Bengio, Clark, Pachocki and more than 20 others, reports AI writing over 80% of approved code at Anthropic and lightly supervised R&D rising from 1% to 26% between March and August. It asks for visibility into lab automation, embedded auditors, capability limits and international agreements.

    No separate Hinton statement on the paper was verified.

    The Next Web

  • Yoshua BengioFounder, LawZero
    Bengio co-authored the intelligence-explosion paper calling for embedded auditors, capability limits and international agreements.The Next WebMoreLess

    The paper's asks go further than the voluntary industry accord signed at the White House the next day.

    The Next Web

  • Jack ClarkCo-founder, Anthropic
    Clark co-authored the paper, which cites Anthropic's own figure that over 80% of its approved code is now written by AI.The Next WebMoreLess

    The data on AI-written code and lightly supervised research comes from inside his company, which makes its publication in a public warning notable.

    The Next Web

  • Jakub PachockiChief scientist, OpenAI
    “the window for action may close”The Next WebMoreLess

    OpenAI's chief scientist co-authored the paper warning that the window for action may close once an intelligence explosion begins.

    It is rare for the person running OpenAI's research to sign a public warning of this kind.

    Not on the roster; added as a guest. No separate statement from him was verified.

    The Next Web

The mood this week

About the same as last week

5 worried4 neutral2 optimistic

Last week

6 worried4 neutral3 optimistic

Worried5 people
  • Yoshua BengioYoshua BengioFounder, LawZeroCapability 5Outlook 2
  • Jack ClarkJack ClarkCo-founder, AnthropicCapability 5Outlook 3
  • Geoffrey HintonGeoffrey HintonAI pioneer, Nobel laureateCapability 5Outlook 2
  • Jakub PachockiJakub PachockiChief scientist, OpenAICapability 5Outlook 2
  • David RobinsonDavid RobinsonFormer OpenAI safety leadCapability 5Outlook 2
Neutral4 people
Optimistic2 people
No new signal7 people

Said nothing new this week, so their scores carry over.

  • Ajeya CotraAjeya CotraMETRCapability 5Outlook 2
  • Clément DelangueClément DelangueCEO, Hugging FaceCapability 4Outlook 3
  • Bill GatesBill GatesChair, Gates FoundationCapability 4Outlook 1
  • Shakeel HashimShakeel HashimEditor, TransformerCapability 5Outlook 3
  • Demis HassabisDemis HassabisCEO, Google DeepMindCapability 5Outlook 3
  • Gary MarcusGary MarcusCognitive scientist and AI criticCapability 3Outlook 2
  • Kevin RooseKevin RooseMachine Gods, NPRCapability 5Outlook 2