CharlieKey takeaways from your favorite podcasts
← 7 November 2026Decoder

How AI is breaking cybersecurity | Decoder

Alex Stamos argues AI has collapsed the skill and cost barriers that sustained the old cybersecurity industry, ushering in a dark age that defenders can shorten only by cutting defense costs, rewriting code, and acting boldly.

Watch on YouTube ↗

Charlie's summary

  1. The kill chain has inverted: AI finds bugs and writes exploits in real time against each target

    Traditionally attackers researched software, found bugs and wrote exploits ahead of time, then a human deployed them. Alex says an AI system can now go hog wild, finding bugs and writing exploits live. He points to RTex, a Chinese open-source tool run on an open-weight model, which hit three South Korean banks and about 200 more companies in a week. It feels like the '90s, he says, except now the whole world runs on the internet.

    ▶ 4:32
  2. The old security industry rested on a skill hierarchy and scarce operators, and AI dissolves both

    Alex says the 2000s-2010s industry assumed tiers of skill, from script kiddies up to a few hundred gods, plus a limited pool of operational staff. Now script kiddies get much of the gods' skill, and two or three Russian kids with Macs or H100s can run many agents all day. Mid-size companies with one security person were already toast against a 20-person ransomware team. Those odds get far worse.

    ▶ 10:07
  3. Defense is an economics problem: attacker token costs go to zero while defenders pay too much

    Alex says attacks are becoming agentic and attackers' cost per token trends toward zero, while defenders' costs stay high. One colleague cited $900 per investigation on a frontier cyber model. Software bug-finding is only 25% of a CISO's job, and the rest needs work too. His answer is tiered triage: cheap classifiers first, then small LLMs escalating to bigger ones. Cognition trains its own open-weight models to cut cost.

    ▶ 16:14
  4. We are entering a dark age, and the job is to shorten it

    Alex is neither a hard doomer nor an optimist. He expects things to get bleaker, with a eventual renaissance that still won't be perfect. Better days arrive as median code is written by LLMs in memory-safe languages. He sees a bad future where each new model brings a fresh pool of superhuman bugs, and the Linux kernel keeps getting more patches after each model release.

    ▶ 21:22

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.