Anthropic's Shift from 'Ban Open Source' to 'Align Open Source': A Deep Dive into AI Safety Paradigms

It's like a three-year-long "gaslighting" session—you clearly remember hearing calls for a "total ban on open-weight models," yet the parties involved now deny it. But thinking back, Anthropic's CEO never explicitly said "we support legislation banning all open-source models" in public. He always said something like "caution is needed for models above certain capability thresholds." This subtle semantic shift caused a massive trust crisis in the tech community, but it also reflects a more fundamental dilemma in AI safety: are we fighting against open source, or against general capabilities that are hard to align?