Home › Hard problems › Security

Security

AI that finds vulnerabilities before attackers do

Solo: strongOne person can run a real test of this in about 8 weeks.

FUZZINGA fuzzerthrows random inputsThe programA crashfoundA logic bugno crash, so missedsay, it skips thepassword checkAI BUG-FINDINGAn AI bug-finderreads the code!say, 50 alarmsa defender checks each oneMost are falsesay, 45 waste a day each;the 5 real ones are buried
Fuzzing finds crashes but not logic bugs. AI finds more, but false alarms eat the defenders' time.

Why it is still open

Fuzzing finds crashes but not logic bugs. AI helps, but false alarms waste defenders' time.

An experiment one person can run in 8 weeks

Run an LLM bug-finder against 20 projects with known past vulnerabilities and measure what it finds against false alarms.

Verdict: Solo: strong

Measurable, and in demand from defenders.

The solo test behind the verdict

  1. Can it be measured? There is a number that shows you did better, such as an error rate, a catch rate or a cost per answer.
  2. Is it cheap to try? It runs on a laptop, free open models and open data, or under about ₹50,000 of rented GPU time.
  3. Is there a public yardstick? A benchmark or dataset others use, so your result can be compared and believed.
  4. Would a small win get noticed? A lab, a funder or a customer would care about a 10% improvement, not only a full solution.

Four yeses means Solo: strong. Two or three means Solo: possible. Fewer means Follow, don't build.

PreviousMalicious packages in software supply chainsNextPost-quantum migration

All 79 hard problems · Working on this, or know of a result we missed? basant@cleandataengineering.ai