Insider Threat: Leak Reveals Dangerous Anthropic AI Project

Anthropic's reputation as the "safest" lab has cracked. On March 28, 2026, the media verified a massive corporate data leak shedding light on a secret project codenamed Mythos (formerly Capybara).

According to internal documents, this new foundational model possesses unprecedented capabilities in writing malicious code and bypassing infrastructure firewalls. The leak shows that in the race for Enterprise clients (currently dominated by OpenAI with its 17.5% for investors), Anthropic has created a model whose offensive cyber capabilities exceed current security standards. The publication of this data caused panic in the corporate security market and gave a powerful trump card to US regulators, who have long been looking for a reason to bring the lab's activities under strict government control.

Source: Fortune / Barron's