Bringing rigour to destruction — pattern never changes: Code implementing protocol gets it wrong, that's your vulnerability site, exploitation is zero-day. Take code everyone trusts, find attack everyone missed.

Britain's most-stolen car, the nineties Vauxhall Nova, no immobiliser, hid an ignition bypass behind a square plastic rocker. Its hazard switch has 180° rotational symmetry: Pull it out, turn it upside down, push it back in, and the cam that merely flashed the hazards now bridges the ignition to twelve-volt battery. The car starts; the key barrel is never touched. Reported to Vauxhall, 2001.

Sometimes protocols are understudied and ill-defined, their security unknown; I write down what they must do to achieve precise security goals. It began in trusted computing's era, with Direct Anonymous Attestation by HP, IBM, and Intel, shipped in billions of chips to prove a machine's capabilities without revealing its identity. I broke its privacy — corrupt administrators, even a passive eavesdropper, could unmask users — pioneered a machine-checkable definition of the privacy it promised, and proved the repaired schemes meet it. Break, define, prove.

Then the web. TLS encrypts a connection but can't force it to finish; truncate it at the right moment and browser & server are left believing different things, a logout that never reaches the server, a warning the browser never sees. I leveraged that into owning Microsoft accounts, partial control of Google's, and throwing elections, on stage at Black Hat'13. And because a standard no one understands is a standard no one secures, I wrote the engineer's guide, raising so many questions IETF revised the standard.

Society operates on critical infrastructure — pipelines, roads, rails spanning the ground beneath us, cargo fleets moving economies through vast oceans, aircraft and satellites up in the sky. "Good" isn't enough; these systems must not fail. They all run on software, mine among them, held to that same unforgiving bar. AI is our rude awakening — this is not a drill — every one of them at brink of collapse.

My agents and I built a battleground. Us against critical infrastructure. Simulating attacks, fortifying live systems, six zero days in six days for a fiver. Unfettered access to classified material (walkthrough'26), remote kill switches (burst'26, shatter'26), kernel vulnerabilities (under disclosure). Day forty-nine: Nineteen zero days disclosed, twenty-one under review, five-hundred plus unopened. Manual triage just collapsed. The patch pipeline is burning out. Summer'26 will witness nations crippled — when your enemy controls your comms, how will you operate?

CISO autonomy, auto-triaging coordinated disclosure into a verified production fix: Cybernuke showcases autonomous detection and protection of content delivery, transport security, and host-to-host comms. Billions spent, decades specifying, auditing, deploying these protocols; nobody checked they're right. They're not.

Three thousand mandatory clauses across twenty-one stacks, hundreds of codebases, tens of languages, several-hundred-million lines of code. Brute force closes enumerable clauses: Test every violation, none land, clause proved. Keys, nonces, signatures put exhaustive search out of reach; handshake shapes so vast Bitcoin's lifetime mining surface is a rounding error. Prove cryptography doesn't matter, only the structure carrying it. Nobody breaks number theory; they break the logic around it. And logic is finite. The challenge is compression into something tractable; we reduce exponential search to linear. Closed provably, not probably.

day zero defence · critical infra patch +1hr · economy patchable on-demand