A one-line 'find PHP vulnerabilities' prompt let a frontier model rediscover a real embargoed flaw
matthew_d_green · x · 2026-10-22
Security researcher Dan Lorenc shared details of the PHP embargo: the same night he learned a PHP vulnerability was pending disclosure, his team independently rediscovered it using their harness and a frontier model with the basic prompt "find bad vulnerabilities in PHP". After getting fully read in, they confirmed findings matched the original report and helped test fixes. When a commenter marveled that a single prompt sufficed, Lorenc confirmed: yes. It's a striking signal that frontier models are now practically useful for real-world vuln discovery.
More from Fun
- "The threat was never rogue AI—it's humans using AI to hurt humans," answered with the 'Always has been' meme — ericwdolan · 2026-10-22
- Dev jokes his job is now just fetching coffee for the security team — jarrodwatts · 2026-10-22
- "Imagine buying a MacBook Pro in 2026" — dev mocks flagship laptop relevance — KevinNaughtonJr · 2026-10-22
- Railway deployment outage drags 16 hours, drawing theo's ire and community jokes — MickeySteamboat · 2026-10-22
- 'I'm considering changing my name to JPEG Morgan' — AI meme of the day — gregmushen · 2026-10-22
- Simpsons meme sums up cybersec's 'worst day of the year — so far' — danielrock · 2026-10-22