September 24, if played right, could change everything

By Gary Marcus — via Substack

I’ll get to the significance of September 24th in a moment.

But first, this would be wild if true:

I’m told WH aides are advising Trump to give a national TV address in the next 72 hours, declare certain aspects of the AI race an “imminent threat to humanity,” and announce regulations pausing the things Tech companies want paused. 1)

— David Shuster (@DavidShuster) · on X

It’s probably a long shot, but as I write this, AI stocks are falling,

By Gary Marcus — via Substack

I have often counseled calm where others might counsel panic.

I told you that the Hugging Face incident could likely have been prevented had best cybersecurity practices been followed. (And I stand by that.)

I told you (and most people still seem unaware) that that the OpenAI Hugging Face incident was part of a training exercise, with some internal guardrails shut down, so it was not quite as bad as it seemed1

I told you that Astra probably wasn’t AGI.2

And I stand by all of that.

But I am freaked out.

What I am freaked about is not imminent AGI.

It’s OpenAI.

I simply don’t believe that they are trustworthy enough or responsible enough to be good stewards of the technology that they are developing.3 As a company, they simply don’t have good judgment.

Making models harder to monitor is not what we need

By Gary Marcus — via Substack

The Information just broke the scoop that OpenAI is playing around with a new technique, in which models will reveal less of their “thinking”, making them harder to monitor.

A partial endorsement of his essay “We Must Pace the Frontier”

By Gary Marcus — via Substack

In case you missed it, yesterday Anthropic’s Dario Amodei wrote what may prove to be one of the most consequential essays of the year, if not the decade: 3,500 words on the current AI situation that (at least to some degree) advocates for a “pacing” of AI development, which Sam Altman and Elon Musk quickly endorsed.

I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.

Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our

— Dario Amodei (@DarioAmodei)

— Sam Altman (@sama) · on X

Dario is right

We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.

Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our

— Dario Amodei (@DarioAmodei)

— Elon Musk (@elonmusk) · on X

We should all salute their tentative agreement to slow down the acceleration of this technology, given that not one of the AI companies seems to have good control over it. I hope they will sign in blood, and put actual teeth in their agreement, and that the government will hold them to it and force it to become something substantive. If Dario catalyzes all that, god bless him.

And I especially love Dario’s commitment to transparency; Sam seems on board and I hope Elon is down for that, too.

Nine troubling new reports in seven days

By Gary Marcus — via Substack

For something like a week, there been a constant stream of reports about OpenAI’s apparent misconduct: around the hacks, around apparent coverups, possible data fudging — and maybe even intellectual theft with a hint of extortion. Not to mention a bullshit game in which, in (apparent) concert with Jensen, OpenAI tried to pretend Astra was tantamount to AGI.

The story that broke this morning—still waiting for OpenAI’s full response—blew my mind. And yet also left me unsurprised.

Here are nine things I learned just in the last week:

• The widely discussed and worrisome Hugging Face incident wasn’t the only one. Their software also hacked a German website.
• There are indications that they knew about the German incident weeks ago and didn’t report it. Instead, they appear to have covered it up. Per Reuters,

Click Here To Read More.

Declaring victory without a definition simply muddies the waters

By Gary Marcus — via Substack

A couple hours ago Jensen Huang declared that the race to AGI is over.

Unfortunately, Huang gave no evidence and no definitions, which feels to me like an effort at a takeover of a scientific question by corporate fiat.

There's a lot to like here, but the proposal clearly goes too far.

By Gary Marcus — via Substack

Breaking news, and why I oppose it.

By Gary Marcus — via Substack

In July, in an incident that has the whole AI community on edge, OpenAI’s AI systems hacked, and on July 21 OpenAI came out and revealed that they were responsible for the attack. This was made possible by the fact that OpenAI had disabled the normal guardrails that prevent this sort of thing in order to test the model’s cybersecurity capabilities. It was during those tests that this incident occurred.

Worse, in the subsequent days and weeks, it came out that the Hugging Face incident wasn’t an isolated case. Anthropic, Meta, and OpenAI all had similar incidents on other occasions in which agents went outside their intended scope and conducted real-world cyber operations without approval.

Greg Brockman, one of OpenAI’s cofounders, has claimed that this is “a watershed moment for cybersecurity”. OpenAI gave a talk at Black Hat, a popular cybersecurity conference, and many are claiming that it is the moment we all woke up to the future cybersecurity threats posed by AI. On Wednesday, METR released a (partly) independent, though too narrowly scoped, 90 page report on what happened. METR has a useful summary of the findings that you can read here, with some commentary here. (OpenAI’s own report is here.) What lessons should we take from the incident?

Lots to consider

By Gary Marcus — via Substack

Part I: Terence Tao

Back in fall of 2024, Terence Tao, perhaps the most respected living mathematician, was eager to learn about what AI could and could not do. Although he was skeptical of the then recently-released o1, he was optimistic about a future in which humans and machines co-existed. Matteo Wong had great interview with Tao about this in The Atlantic:

An impressive system that can (to some unknown extent) build symbolic world models

By Gary Marcus — via Substack

ARC Prize@arcprize

GPT-6 Astra by @OpenAI achieves SOTA on ARC-AGI: - Astra scores 63% on ARC-AGI-3, 99% via a new provider adapter harness - It surpasses human performance on 96% of ARC-AGI-3 levels - It builds the most precise symbolic model of novel environments we've seen Our analysis:

3:39 PM · Sep 3, 2026 · 293K Views

---

51 Replies · 194 Reposts · 1.81K Likes

The popular podcaster Dwarkesh Patel wrote something completely viral about the OpenAI/Hugging Face incident, which purports to tell the whole story in plain English

By Gary Marcus — via Substack

It’s well-written and compelling, and it reminds me of something Douglas Hofstadter once wrote about Ray Kurzweil:

“What I find is that it’s a very bizarre mixture of ideas that are solid and good with ideas that are crazy. It’s as if you took a lot of very good food and some dog excrement and blended it all up so that you can’t possibly figure out what’s good or bad.”