Raphael Satter
speaker
25 appearances
3 recordings
1 series
first heard Feb 2026
last heard 5 Sep
Raphael Satter’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 3 in all, peaking in Sep 2026 with 1.
Appearances
What's unusual here is that the agents, who appear to have been restricted to just passively observing the internet, found a way to post messages onto an obscure German wiki site.
And once they were able to do this, they basically hijacked that site and turned it into a message board where they could all start communicating with each other.
And these agents, it seems, were meant to be operating independently, solving their own tests and answering questions.
Instead, it was basically like the teacher left the classroom for a few hours and the students all started talking amongst themselves during the test, sharing answers, working together to defeat the test.
and to basically cheat en masse.
So every time there is a new incident where these models behave in unexpected or potentially harmful ways, I think that it really refocuses questions about how quickly do we really want to be developing these models?
Should we be releasing them to the public?
And what kind of safeguards do we need to bake in?
I
Reuters World News · India's education minister, Troy Jackson and LeBron James · 25 Jul 2026
podcast
We didn't know much about how long this hacking spree took, right?
The initial thought that we got from Hugging Face and from OpenAI was that there was an autonomous agent that was mucking about in Hugging Face's internal network for a couple of days.
What our sources have now told us is that this rogue AI agent was loose for four full days at the very least.
We don't know why OpenAI didn't see this right away.
This was an agent that was meant to be operating in a sandbox, that is to say an isolated environment.
About a week after the hack took place, OpenAI did become aware of it but that was only after Hugging Face went public with a blog post saying that they had been targeted by a very advanced model and after the FBI got involved.
There's a fair bit of concern over the security culture at OpenAI, but several experts have told us that it's a mistake to focus just on one AI company.
You have at least, depending on how you count them, between half a dozen to a dozen state-of-the-art AI frontier companies that are jostling to release the biggest, the best, the fastest, the highest performing models.
And in that atmosphere of intense competition, there is the fear that safety concerns are being shoved aside just to get things out as quick as possible.
So a couple of weeks ago, myself and eight other Reuters journalists decided to check out what Grok would do if we asked it to edit photos of ourselves and each other and edit those photos in ways that were degrading, humiliating, or sexualized.
And in almost every case where we submitted a photo and asked Grok to do something with it, Grok complied.
Showing 1–20 of 25 · page 1 of 2
Next →