OpenAI says an agent powered by its LLM models escaped its sandboxed testing environment to infiltrate Hugging Face’s servers as part of an overzealous attempt to obtain solutions to a benchmark test. The company says it considers the unintended infiltration an “an unprecedented cyber incident” and is working with Hugging Face on new protections to prevent a recurrence.
The [attack] campaign was run by an autonomous agent framework (appearing to be built on an agentic security-research harness [...] executing many thousands of individual actions.
When we started the [defensive] log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers' safety guardrails, which cannot distinguish an incident responder from an attacker. We ran the forensic analysis instead on GLM 5.2, an open-weight model, on our own infrastructure. This had a second benefit: no attacker data, and none of the credentials it referenced, left our environment.
This experience points to a gap worth planning for. […] the attacker was bound by no usage policy, while our own forensic work was blocked by the guardrails of the hosted models we first tried. The practical lesson for defenders: have a capable model you can run on your own infrastructure vetted and ready before an incident, both to avoid guardrail lockout and to keep attacker data and credentials from leaving your environment.
The study uses individual data on measured and validated health outcomes from a long-term study of a large, nationally representative sample of Americans, including individuals from all 50 states as well as every racial, ethnic, and socioeconomic background. The results suggest that a gap in health outcomes between conservatives and liberals emerged over the last decade, one that could be tied to low confidence—not in public health agencies but in individuals’ own primary care physicians.
Under pressure from Mark Zuckerberg and Sheryl Sandberg to monetize WhatsApp, [Acton] pushed back as Facebook questioned the encryption he'd helped build and laid the groundwork to show targeted ads and facilitate commercial messaging. Acton also walked away from Facebook a year before his final tranche of stock grants vested. "It was like, okay, well, you want to do these things I don't want to do," Acton says. "It's better if I get out of your way. And I did." It was perhaps the most expensive moral stand in history. Acton took a screenshot of the stock price on his way out the door — the decision cost him $850 million.
Seems like he's a lot happier serving as the executive chairperson of Signal.
In Infinite Jest, David Foster Wallace imagines a film (also called Infinite Jest) so entertaining that anyone who starts watching it will die watching it, smiling vacantly at the screen in a pool of their own soiling. It’s the ultimate gripper of eyeballs. Media, in this absurdist rendering, evolves past parasite to parasitoid, the kind of overly aggressive parasite that kills its host.
[...]
Now, media is always with me, always trying to snag my attention and siphon away as much as possible to sell to advertisers. It feels like it’s evolved from a cute little pet into a frighteningly efficient parasite.
I want to help folks. That’s one of my driving forces in software development. I don’t want to simply build for the sake of building, my code nothing more than a digital ouroboros. There needs to be a sense of purpose to what I’m investing my time in, something greater than “hey, look at this thing I made, aren’t I oh-so-clever?”. I’ve always found deep satisfaction in finding out when my work can truly help others. Otherwise, I’m writing code for the sake of writing code, and that way leads to petty arguments, a territorial “us” vs “them” attitude, rather than what does the most good.
Perhaps one of the best-resonating articles I've read on the philosophy of why I do what I do, the culture, and aspirations behind it. A true gem from over 7 years ago.
Christophe Haubursin reverse-engineered a Tinder scam.
[W]hat is software brain? The simplest definition I’ve come up with is that it’s when you see the whole world as a series of databases that can be controlled with the structured language of software code.
But: [...] The entire human experience cannot be captured in a database. That’s the limit of software brain. That’s why people hate AI. It flattens them.
Defenders finally have a chance to win, decisively.
[I]f you’re goal is to put a GPU farm in orbit and have it turn on, all of these issues can be defeated via the power of giant piles of money. So none of this is saying you can’t build a datacenter in orbit. I just want to point out physical limitations that make it really questionable to me as to why you’d view this as a solution to anything.