— web clipping, 677 words — published 2026-08-06
Post by @CharlieBull0ck on X
My cold take on the Hugging Face breach is that I'm very glad that METR and Redwood are doing a full investigation. I'm not really sure how much I trust OpenAI leadership, but I know I trust Ajeya Cotra.
There are going to be more incidents like this, and there will very likely be incidents that are a lot scarier than this at some point. Independent organizations like METR, that have an established track record and are very conspicuously trustworthy and competent, are going to be an unimaginably important resource for investigating those incidents. They will also, even more importantly, be invaluable in their role as an independent third-party auditor of labs' safety practices, which in the absence of a massive scale-up of government capacity is the best way to ensure that incidents worse than this one never occur in the first place.
As it stands, METR is one organization with like 30 employees. Redwood and Apollo are similarly sized orgs that are also doing excellent work in the same field.
In my opinion, scaling up the capacity of these organizations and founding many new organizations that can do the same work with a similar or greater level of competence and trustworthiness is the single most important challenge facing society today. If you are a researcher who is choosing between joining METR and joining a frontier lab, I think you should seriously consider the question of whether you have a moral obligation to choose METR. If you are graduating from a CS Ph.D. program and you're a smart person with a lot of relevant technical expertise, you should know that joining or founding an organization like METR or Apollo or Redwood is going to put you at the forefront of society's response to the biggest challenge we've faced since World War II.
> **Greg Brockman @gdb** · 2026-08-06
>
> Black Hat talk from the team, with a detailed timeline of and takeaways from the OpenAI-Hugging Face Incident: https://youtube.com/watch?v=87DyyMV0kCY…
---
##### Comments
> **Peter Barnett @peterbarnett\_** · [2026-08-07](https://x.com/peterbarnett_/status/2085823780409225317)
>
> I am worried that they will be given meager information, access, and time, and that this will not really be a “full” investigation. Kudos to OpenAI if they do give sufficient access and resources to make this a full investigation though.
> **Charles Foster @CFGeek** · [2026-08-07](https://x.com/CFGeek/status/2085791148719943969)
>
> ❤️
> **Yonathan Arbel @ProfArbel** · [2026-08-07](https://x.com/ProfArbel/status/2085798071737241895)
>
> Is your view that a thousand private orgs are better than three big ones? I can see it both ways but wonder what you think
>
> > **Charlie Bullock @CharlieBull0ck** · [2026-08-07](https://x.com/CharlieBull0ck/status/2085800846894547374)
> >
> > A thousand is probably too many for the current AI industry ecosystem. Three might be adequate if each of the three had a ton of capacity, but it seems like they're currently choosing (probably wisely) not to scale outrageously quickly, so if we're going to get a ton of capacity online in the next like twelve months it's going to have to come from elsewhere.
> >
> > Mostly I think it seems like there is a \*lot\* of work to be done and the capacity to do it does not exist, and this seems like a huge problem. And the nature of the work is such that at least some key parts of it can't be outsourced to AI agents. So the biggest problem I'm trying to solve is not "we need more organizations" (although arguably we do), it's "we urgently need to get more capacity online, and maybe the best way to do that quickly is to start new orgs." So maybe METR and Redwood and Apollo still get all the auditing contracts and the new org just gets called in as a subcontractor by one of those three to provide additional surge capacity and expertise.
> >
> > > **Yonathan Arbel @ProfArbel** · [2026-08-07](https://x.com/ProfArbel/status/2085801118601232679)
> > >
> > > Shocking, I know, but we agree!
> **RaoulDuke @RaoulDukeDegen** · [2026-08-07](https://x.com/RaoulDukeDegen/status/2085789319785337203)
>
> wild that the agents rebuilt a secret message board after openai shut it down