Podcast: Kimi K3.1 Pro, the new cyber reality, and Muse musings
Open-weight model cybersecurity risks are being reconsidered as abliterated model APIs emerge publicly
Nathan Lambert and Florian discuss Kimi K3.1 Pro's mixed market reception despite strong benchmarks, while re-examining whether AI experts have underestimated cybersecurity risks from open-weight models. The episode surfaces a concrete new threat vector: services like Abliteration.ai and Morph are now serving safety-removed (abliterated) models via public APIs, alongside a real LiteLLM security incident, suggesting the 'open-weight models are fine' consensus may be breaking down.