antirez releases steering vector file to make LLMs comply with cybersecurity work
antirez · x · 2026-09-03
Redis creator antirez argues that asymmetric access to LLMs for cybersecurity work is a real problem, and has released a steering vector file that can be used with his DwarfStar tool.
- Usage: load via --dir-steering-file <file> and --dist-steering-ffn flags, trying strengths 2, 3, or 4 depending on how much the model refuses
- Goal: make the DS4F model cooperate on cybersecurity tasks instead of triggering refusal guardrails
- He followed up with the actual file download link
The move is a concrete example of using activation steering to alter model refusal behavior, feeding into the ongoing debate about asymmetric LLM access on offense vs. defense in security.
Related event: antirez Releases Steering Vector to Bypass LLM Refusals(4 posts)→
More from Models
- Grok goes down with barely a ripple on the timeline — toptickcrypto · 2026-09-03
- TabICLv2 talk advancing open tabular foundation models now public from ICML workshop — RichmanRonald · 2026-09-03
- Yacine MTB: Everything is an RL env — publish one for your task and let models overfit it — yacineMTB · 2026-09-03
- Grok Build and Grok 4.6 web UI go down at once, fueling speculation of a Colossus 1 outage — PawelHuryn · 2026-09-03
- As Claude Goes Down, X User Jokes OpenAI Should Seize the Moment With a Surprise 'Astra' Release — kimmonismus · 2026-09-03
- User ditches GPT-5.6-sol for DeepSeek V4 Pro: 'feels like Opus 7.0 landed' — ryunuck · 2026-09-03