Never let a crisis go to waste : how we used GenAI to battle a SPAM attack
summary
A small dev team at [large publishing company] faces a coordinated attack that floods the preprint server with ~2,000 phishing submissions per day and overwhelms human reviewers; firewall, CAPTCHA and regex defences reach ~80% and keep breaking. The team ships a zero-shot prompt calling an internal GPT4.1 and integrates it into the event-driven pipeline on Java on Kubernetes, keeps a human-in-the-loop, launches in three weeks, and achieves about 95% accuracy while costing roughly $200/month.
AI-written summary · about
