Anthropic faces backlash over Claude Mythos claims

Anthropic is accused of using "fear-selling" and manufactured scarcity to position itself as the gatekeeper of a private, unaccountable security cartel.

• –Firefox exploit success rates dropped from 72% to 4% when modern security patches were re-enabled, suggesting Mythos fails against defended systems.
• –The $100M Project Glasswing defensive fund is mostly usage credits for Mythos itself, effectively subsidizing partners to validate Anthropic's own product.
• –Small open-weights models (3.6B) can reproduce the "showcase" bugs for pennies, undermining the claim that these are exclusive frontier capabilities.
• –The lack of standard metrics like CVE lists or disclosure timelines in the 244-page system card signals a PR play rather than a scientific breakthrough.
• –Restricting access to incumbents creates a moat under the guise of safety, a classic move for regulatory capture in the burgeoning AI market.