AI Just Built a Cyberweapon. Is Anyone Ready?
Anthropic's new Mythos model didn't just get better at writing code — it got better at breaking it. In an hour, an AI mapped decades of hidden vulnerabilities across live systems. In four hours, a supply chain attack silently exfiltrated 500,000 credentials and compromised 20,000 repositories. The question isn't whether this is alarming. It's whether the companies and governments responsible for protecting critical infrastructure — water, power, gas — are anywhere close to ready. On this episode of The BroBots, Jeremy and Jason work through what Anthropic's internal memo actually said, what a cyberweapon-grade AI changes about the attack surface, and why Jason thinks the survivalists have been right all along.
Key Moments
- 00:00 — Anthropic's Mythos: what the internal memo actually said and why it's different
- 01:38 — The LiteLLM supply chain attack: how 500,000 credentials were stolen in 4 hours
- 04:27 — Zero-day attacks explained: why signature-based detection can't stop what it hasn't seen
- 06:44 — Mythos vs. prior models: from 60s to 77–78% effectiveness — what that jump means
- 09:34 — Jeremy tries to find the optimism: Glasswing, the $100M security head start
- 11:06 — The real threat: why utilities and infrastructure are the soft targets
- 13:21 — Regulation vs. arms race: should billionaire AI companies have a leash?
- 15:13 — The nuclear analogy: what a global AI treaty would actually require
- 17:09 — 'Easy mode': the counterargument that Mythos's test conditions were unrealistic
- 20:22 — Jason's actual survival advice: fire, water, neighbors, and a CRT in the attic
Follow Brobots: www.brobots.me/follow
