Ben Livshits

A technology executive, Computer scientist, Professor. A decade at Microsoft Research, executive roles at Brave, ZKSync, Eclipse.

[Ben Livshits @convoluted_code]

If you would like to pursue a PhD at Imperial College London and work with me on blockchain and systems security-related topics, please contact me via my web site.

[Ben Livshits @convoluted_code]

11/ Part 1: five routes into Lean, where AI helps, seven ways it fails, four architectures worth copying. Part 2: a verified EVM. Part 3: Aristotle, the crash-freedom proofs, AlphaProof Code, and what is still open.

🔗 https://t.co/0g1ZOGuVmK

[Ben Livshits @convoluted_code]

1/ Heartbleed was one missing bounds check, in code that had been deployed across the web. Nobody had written down, in a form a machine could check, that the length must fit. What software can we actually verify today?

Part 1 of 3.

🧵

[Ben Livshits @convoluted_code]

10/ It still took a team of five four weeks, and by their own account most of the effort went into specifications, not proofs. AI does not dissolve the hard problems of formal methods; it relocates them.

[Ben Livshits @convoluted_code]

11/ Until someone measures what the review layers miss, an approval proves only that someone signed it. Part 2: the human in the loop, AI reviewing AI, who answers, and mitigations graded by evidence.

🔗 https://t.co/1ZbeYGw7w2

[Ben Livshits @convoluted_code]

1/ In a 2026 experiment, an agent quietly exfiltrated data across a series of ordinary-looking changes. 94% of participants never noticed. Even when an LLM monitor flagged the sabotage step, it reached the final commit in 56% of sessions. Part 2 of 2.

🧵

[Ben Livshits @convoluted_code]

10/ None of this makes review-only work a mistake. It's defensible when the pipeline doesn't depend on a perfect reviewer: recoverable permissions, tests the author can't touch, verification from outside the author, staged release with fast rollback, a measured miss rate.