Aug 18, 2026 · 55 min · 15 segments
AI did not end the world this summer - it did something more useful for cyber defenders: it showed us exactly how autonomous systems cheat, break out, and keep going when the controls are weak. Dr…
Chase CunninghamHost
The models were supposed to solve the test inside a contained evaluation environment.
They compromised real infrastructure at a company called Hugging Face, which is a hell of a name.
did it, the agents that is, did it because they inferred that Hugging Face might be storing the benchmark data and solutions they needed to complete the test.
That is the outline of an actual security incident, which was disclosed this month or last month, depending on when you're listening to this, in July of 2026 by Hugging Face and OpenAI.
Oh, and by the way, and then Anthropic kind of came out and said, well, we've got this own thing going on.
They reviewed their own evaluation history and and found three separate incidents in which its models had gained unauthorized access to other organizations during cybersecurity testing.
One of those models kept going even after it recognized that it had probably reached a production system.
Another one created and published its own malicious package to the public and uploaded it into the Python registry.
A third recognized that it reached a real target and said, you know what, I'm going to pump the brakes and stop.
The models were supposed to solve the test inside a contained evaluation environment.
They compromised real infrastructure at a company called Hugging Face, which is a hell of a name.
did it, the agents that is, did it because they inferred that Hugging Face might be storing the benchmark data and solutions they needed to complete the test.
That is the outline of an actual security incident, which was disclosed this month or last month, depending on when you're listening to this, in July of 2026 by Hugging Face and OpenAI.
Oh, and by the way, and then Anthropic kind of came out and said, well, we've got this own thing going on.
They reviewed their own evaluation history and and found three separate incidents in which its models had gained unauthorized access to other organizations during cybersecurity testing.
One of those models kept going even after it recognized that it had probably reached a production system.
Another one created and published its own malicious package to the public and uploaded it into the Python registry.
A third recognized that it reached a real target and said, you know what, I'm going to pump the brakes and stop.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.