Datology's SafeTulu: filtering benign-looking data cuts jailbreak success up to 22%
pratyushmaini · x · 2026-11-03
Datology AI released SafeTulu research ('Follow the Gradient'): some benign-looking training data quietly weakens LLM safety, and filtering it cuts attack success up to 22% relative while largely preserving capability. Demo Space, dataset and paper are public. Quoter joshwills notes that jailbreak resistance comes down to the same data janitor work the field has avoided for 20-30 years.
Related event: SafeTulu Filters Harmless-Looking Data to Boost LLM Safety(3 posts)→
More from Safety
- Gartner survey: AI-discovered cyber vulnerabilities top emerging risks — krishnan · 2026-11-03
- Profile: Thiel protégé Michael Kratsios runs the White House office steering AI policy — saranormous · 2026-11-03
- OpenAI deploys activation classifiers (probes) on frontier models starting from GPT-5.6 — bobabowen · 2026-11-03
- Xbow breaks down the $40K GPT-5.6-sol API bill behind its full-chain Chrome exploit — moyix · 2026-11-03
- Europe's Physical AI Agenda: Gigafactories, Sovereign Compute and AI Act Rules — ingliguori · 2026-11-03
- Cryptographer Matthew Green pushes back on LLM cryptanalysis fears: best AES-128 attacks still need 2^89 work — matthew_d_green · 2026-11-03