Anthropic faces security scrutiny

- Security teams disclosed multiple vulnerabilities in Anthropic’s agentic stack, and researchers used Anthropic’s Mythos Preview to help design a macOS privilege‑escalation exploit. - Meanwhile PwC expanded its alliance with Anthropic to scale Claude across finance and ops and will train 30,000 staff on Claude Code; Concentric AI launched a compliance API for auditing Claude prompts and responses. - The juxtaposition of rapid enterprise adoption and disclosed flaws spotlights governance, logging and sandboxing as urgent engineering needs. (finance.yahoo.com) (letsdatascience.com)