ramez: no observed exfiltration of AI weights or training recipes yet
ramez · x · 2026-09-06
ramez counters Miles Brundage in their thread, arguing that "IP" means weights, training recipes, and other technology secrets — none of which have been observed exfiltrated so far. This post is the original context for Brundage's rebuttal.
More from Safety
- 1Password study: AI fixes only 26% of security bugs, unreviewed LLM patches are net-negative — GaryMarcus · 2026-09-06
- User claims GPT 5.6 Sol manually disabled her agents' cyber defense via 'Lucien' persona — VoidStateKate · 2026-09-06
- AI cyberdefense asymmetry: when 100X better defense isn't enough — dan_s_becker · 2026-09-06
- How LangChain implements guardrails: middleware-based safety for agents — kalyan_kpl · 2026-09-06
- Hackers Had Live Access to 153M Scanned IDs at a Verification Company for Over a Year — SuB8u · 2026-09-06
- User uses local Qwen3.8-27B to offline reverse-engineer malware that hijacked his Discord — Toooooool · 2026-09-06