Anthropic ships two models sharing core architecture

The "two passports" approach indicates a nuanced release strategy from Anthropic, likely balancing different regulatory environments, user tiers, or safety thresholds with the same underlying model weights. Delaying a model release specifically because it "got too good at hacking" is a strong signal about the current frontier of AI capabilities and the very real security concerns labs are facing. Shipping two variants of the "same brain" might become a standard industry practice to serve both enterprise/secure environments and general public use without training entirely separate foundational models.