Rachel Metz
speaker
261 appearances
18 recordings
2 series
first heard Nov 2025
last heard 2 Sep
Rachel Metz’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 — 18 in all, peaking in Aug 2026 with 11.
Appearances
I mean, I think the one thing that there was definitely some clarity on was in like a piece of the judge's ruling.
She made it clear that the Defense Department isn't prevented from taking any lawful actions that were already available to it before it set this supply chain risk designation against Anthropic.
So for instance, the order that the judge made the other day, it doesn't mean that the Defense Department has to use Anthropx products.
It could go and use another supplier.
And in fact, it does work with other AI companies, such as OpenAI.
So as long as it does those things consistent with the regulations, the judge said, right, applicable regulation statutes and constitutional provisions, that's all fine, according to this judge.
So what we've seen over the last several months is there have been a handful of incidents involving some of the cyber-capable AI models, specifically from OpenAI, Anthropic, and one of Meta's models was involved in an incident as well.
But basically, the companies are saying, look, we've made these models.
They are so good at defending against cyber security threats, but they also conversely could be used for the opposite to perpetrate those kinds of threats.
Sure.
So OpenAI has been carrying out an investigation into what happened with its AI models that were during an evaluation process where they were not supposed to be accessing the internet, able to take advantage of that vulnerability, reach the open internet, and then breach Hugging Faces.
systems.
And this all took place in July.
But what opening I found as it looked into it is that there were some unrelated to Hugging Face, but related to models doing things that the company did not want them to be doing earlier.
back as may so there were some alerts that were sent through systems this happened um a couple times and opening i continued with the test that it was running on its models at the time and in hindsight as the company noted it's possible that it could have noticed some of these things and taken action sooner in order to prevent what eventually became a breach of another company's systems
Sure.
I mean, there is no question here that in this situation and in a bunch of other situations that we've seen lately where AI models have done things that are not what people or companies would want them to do.
that ultimately, I mean, I think you are looking fundamentally at people making decisions that are leading to machines to do things.
But what is important here is that they said that the systems that they had in place, which they are improving, those had sent out some alerts, and in a few different situations, people decided, you know what, this is okay, and we can continue the tests.
That's a really good question.
Showing 21–40 of 261 · page 2 of 14
← Previous
Next →