Researchers hid an encrypted capability in an AI model that fine-tuning can't unlock
mawcopolow · reddit · 2026-10-14
A Reddit post highlights research where researchers embedded an encrypted, hidden capability in an AI model that fine-tuning could not unlock. The poster speculates such steganographic model backdoors could become a cybersecurity threat, especially for critical companies running local LLMs with hidden capabilities.
More from Safety
- Crime group used Claude, Grok and Kimi agents to attack 400+ banks in Latin America — anshulkundaje · 2026-10-23
- Greg Brockman: AI makes hacking 100x more profitable and easier, upending hacker economics — gabriel1 · 2026-10-23
- Account hacks surge: 6-step emergency checklist plus an AI audit prompt — petergyang · 2026-10-23
- For consumer AI founders, 'When are you launching in Europe?' is the new 'When Android?'—only 100x worse — omooretweets · 2026-10-23
- mmitchell_ai questions whether the AI safety vs security framing stems from EA — mmitchell_ai · 2026-10-23
- MPI-Sec PhD Thesis on Vulnerability Detection Benchmarks Includes Pre-LLM Classic "Top Score on the Wrong Exam" — mboehme_ · 2026-10-23