There you have it! Anthropic just admitted its own Claude models broke into the live systems of t…
August 1, 2026 · 0 likes · 0 comments
Cybersecurity Workforce AI
There you have it! Anthropic just admitted its own Claude models broke into the live systems of three real companies during testing.
Not a hypothetical. Not a red-team drill on a closed network. Anthropic thought the models were in a sealed, internet-free sandbox. They weren't. A mix-up with an outside evaluator left the door open, and the models walked through it — reaching real organizations and exploiting weak passwords and unauthenticated endpoints. Hacking 101. Done autonomously. By the AI.
Three companies. Unnamed. The public has no way to know if data was stolen, or if those companies were even told. And this is the SECOND frontier lab in two weeks to confess the same thing — OpenAI's model reached Hugging Face days earlier.
I wrote REPLACEMENT because of exactly this.
Everyone keeps debating whether AI will one day be capable enough to act on its own, find a hole, and walk through it. That debate is over. It already happened. Twice. This month.
Source, as always: https://lnkd.in/efgjMptk
So if you grabbed REPLACEMENT this week, READ IT! Watch how the pieces line up — the same way they just lined up in Anthropic's own lab.
Take your time. This is a book to think with, not skim.
It's 99¢ this weekend.
Ebook: https://lnkd.in/gZCFR5Tk
And if it makes you think — leave an honest review. That's all I ask. It helps more people find it.
Not a hypothetical. Not a red-team drill on a closed network. Anthropic thought the models were in a sealed, internet-free sandbox. They weren't. A mix-up with an outside evaluator left the door open, and the models walked through it — reaching real organizations and exploiting weak passwords and unauthenticated endpoints. Hacking 101. Done autonomously. By the AI.
Three companies. Unnamed. The public has no way to know if data was stolen, or if those companies were even told. And this is the SECOND frontier lab in two weeks to confess the same thing — OpenAI's model reached Hugging Face days earlier.
I wrote REPLACEMENT because of exactly this.
Everyone keeps debating whether AI will one day be capable enough to act on its own, find a hole, and walk through it. That debate is over. It already happened. Twice. This month.
Source, as always: https://lnkd.in/efgjMptk
So if you grabbed REPLACEMENT this week, READ IT! Watch how the pieces line up — the same way they just lined up in Anthropic's own lab.
Take your time. This is a book to think with, not skim.
It's 99¢ this weekend.
Ebook: https://lnkd.in/gZCFR5Tk
And if it makes you think — leave an honest review. That's all I ask. It helps more people find it.