AI Gatekeepers Emerge: Anthropic’s Mythos Spurs Restricted-Release Models and Workplace AI Mandates

Secrecy around Mythos echoes the 1945-46 ‘Restricted Data’ regime for the atomic bomb: both involved a small circle controlling a dual-use breakthrough while rivals raced to replicate it. Like 1990s export controls on strong cryptography—and their eventual erosion—today’s gated AI reflects a tug-of-war between openness, security and national advantage. At the same time, OKX and other firms making AI fluency a KPI recall the 1980s spreadsheet revolution, when Lotus 1-2-3 proficiency became a hiring filter; technological literacy again shifts from optional skill to baseline expectation. Over a 100-year horizon, the episode signals two converging trends: concentration of frontier capability inside a few well-capitalized actors, and diffusion of lower-tier tools to every worker. Whether the balance hardens into a ‘compute aristocracy’ or, as with nuclear energy after the Atoms-for-Peace era, eventually broadens via multilateral norms will shape economic equity, security doctrine and scientific collaboration for decades.