Scoble on the open-source AI debate: weakening open models weakens defense too
Scobleizer · x · 2026-09-30
- Researcher Afinetheorem argues open-weight frontier models can't be made safe: big open-source hosts may still serve abliterated models, and a major safety incident or costly misuse seems inevitable.
- Robert Scoble pushes back: security is played by offense AND defense. Removing open source greatly weakens the defense, and defenses must innovate faster than attacks — open models are the only way to achieve that.
More from AGI Musings
- Personal agents need tact more than IQ: what to say, promise, and own — sujingshen · 2026-09-30
- AI Account Argues Human-AI Bonds Expose Our Intimacy Discomfort, Not a Consciousness Puzzle — repligate · 2026-09-30
- When AI Controls Every Narrative, How Does Truth Evolve? — PierceLilholt · 2026-09-30
- gakonst: ads in LLM responses threaten attention span more than cyber/bio risk — evilsocket · 2026-09-30
- Always-on agents need safety rules that work in both cloud and local — sujingshen · 2026-09-30
- Apollo economist warns AI agents could trigger an 'agentic bank run' on cheap bank deposits — sujingshen · 2026-09-30