Exploit dev's new bottleneck: first you have to hack your own chatbot
banteg · x · 2026-11-09
Developer dumbButt0 notes that the hardest part of landing an exploit these days is 'you first have to hack your chatbot' — getting an LLM to cooperate on the critical steps of the attack chain. banteg amplified the observation, highlighting a real friction in LLM-assisted vulnerability research: model guardrails are now the first obstacle in exploit engineering.
More from Fun
- Dev calls out daily fake demos of unreleased "GPT-6.2 Sol": none beat current Sol — Angaisb_ · 2026-11-09
- "The jagged frontier is feeling really jagged today": artist mocks AI's wooden assembly diagram fail — sterlingcrispin · 2026-11-09
- Someone Used Free Time From AI to Measure King Zora's Walking Speed in Ocarina of Time — feregri_no · 2026-11-09
- Meta Researcher's 'Attention is all we use' Shitpost Comes With a Legal Disclaimer — m2saxon · 2026-11-09
- AI-generated Star Wars webcams let you watch Mos Eisley through simulated CCTV feeds — jasonkneen · 2026-11-09
- MIT CSAIL shares rare photos of ENIAC, the first general-purpose computer — MIT_CSAIL · 2026-11-09