Lambert says open-weight models add cyber risk but are not a 100x multiplier

AINathan Lambert says open models will be used for cyberattacks, as closed models already have been. He argues they are not a leverage tool that increases cybercrime 100x, because the bottlenecks lie elsewhere.

AINathan Lambert says he was not wrong about open models and cyber risks, arguing the AI community focuses on which model type causes harm rather than how cybersecurity maintains equilibrium, who enacts harm, and how they use new tools. He develops the argument in a post on Interconnects.

AINathan Lambert says he now doubts his earlier claim that an open-weight release of Claude Mythos would be a manageable step rather than a societal threat. He cites warnings from Rob Wiblin, Anthropic's Project Glasswing post, and Zvi Mowshowitz that call Mythos an actual disaster or chaos if misused. Lambert says Epoch AI used GLM-5.3 to find a Firefox exploit, but that a working attack needs several stages to align.