Cybersecurity Looks Like Proof of Work Now
April 15, 2026
Last week we learned about Anthropic’s Mythos, a new LLM so “strikingly capable at computer security tasks” that Anthropic didn’t release it publicly. Instead, only critical software makers have been granted access, providing them time to harden their systems.
This chart suggests an interesting security economy: to harden a system we need to spend more tokens discovering exploits than attackers spend exploiting them.
While there has been skepticism, to say the least from some quarters as to the claims about Anthropic’s new Mythos models capabilities when it comes to cybersecurity, third-party analysis is now suggesting that it does have capabilities we’ve yet to have seen in large language models to date.
Here Drew Brunick makes observation that bad actors are now only bounded by money and that in the ongoing Red Queen dilemma of cybersecurity the only response Istio spend more money more quickly on tokens defending against vulnerabilities in your software and the software that you rely on.
