Mythos forces South African banks onto high alert

Capitec CEO Graham Lee has confirmed the bank applied for early access to Anthropic's Claude Mythos model, citing rapidly shrinking vulnerability patch windows as a key concern. Mythos, restricted to roughly 40 organisations via Anthropic's Project Glasswing, has been flagged by the British AI Security Institute as substantially more capable at cyber offence than any previously assessed model. South African banks including Standard Bank are monitoring developments, while major US banks like JPMorgan already have access. Security experts warn that Mythos marks the shift of AI-powered vulnerability discovery from theoretical to operational threat, with agentic AI enabling attackers to exploit weaknesses at nation-state-level speed and scale. Anthropic's asymmetric early-access strategy is designed to give defenders a head-start, though institutions without access face a growing security and competitive gap.