SAN FRANCISCO, California — Anthropic an­nounced Tuesday that its newest model, Claude Mythos, is too dan­ger­ous for im­me­di­ate pub­lic re­lease af­ter in­ter­nal test­ing found it could “meaningfully dis­rupt in­for­ma­tion ecosys­tems and in­sti­tu­tional de­ci­sion-mak­ing at scale.” The com­pany said it will in­stead launch the sys­tem through a re­stricted en­ter­prise pro­gram for se­lect Fortune 500 part­ners.

Anthropic de­scribed the move as a re­spon­si­ble de­ploy­ment strat­egy. While re­leas­ing a civ­i­liza­tion desta­bi­liz­ing model to the gen­eral pub­lic could cre­ate un­pre­dictable so­ci­etal risks, the com­pany noted that large multi­na­tional firms al­ready have deep ex­pe­ri­ence caus­ing so­ci­etal dis­rup­tion in highly struc­tured en­vi­ron­ments. According to Anthropic, en­ter­prise cus­tomers of­fer a safer set­ting be­cause any re­sult­ing harms can be tracked through quar­terly re­port­ing, com­pli­ance work­flows, and in­vestor dis­clo­sures.

The com­pany said Mythos will be re­leased to the pub­lic once its le­gal team fin­ishes up­dat­ing its terms of ser­vice to clar­ify that any harms as­so­ci­ated with so­ci­etal col­lapse are not el­i­gi­ble for class-ac­tion treat­ment.