5 Comments
User's avatar
Travis Wright's avatar

The detail that jumped out is attackers stealing API keys for inference instead of mining crypto, because that quietly turns every agent fleet's key file into the new hot wallet. Once that's true, agent identity and per-key spend caps stop being nice-to-haves and start being the firewall. Also, running OpenBSD as your desktop in high school is the most credible security credential I've ever seen in a byline.

Ollie Darby's avatar

the point about enterprises holding off on auto updates because breaking things felt riskier than not patching is so true, and if offense really starts running at machine speed that tradeoff flips pretty quickly. The patching startups feel like the less glamorous but more obvious bet to me, especially with how much software inside a big company isn't even maintained anymore

Tanya Trushnikova's avatar

Companies are so used to playing it safe that they sometimes forget doing nothing can be just as risky. And somehow, it's still easier to explain why you didn't change anything than why you did.

Nan Deng's avatar

The Windows XP "surviving the first day" analogy is the part that feels current: if exploit time has collapsed but most packages still wait for a cautious release window, the unpatched interval is the product. I'm curious whether the inference-key theft they flag changes how teams store those keys — because a worm that only needs a valid key can skip the old "knock the network offline" behavior entirely.

Aakash Gupta's avatar

Supply chain risk lands on whoever approves the packages their Claude Code agent picks