XNews

Anthropic Research and Security Findings

Anthropic-related posts cover a red-team blockchain attack and an internal AI usage review. - Anthropic red team used AI to attack blockchain and reportedly stole $4.6M - The research asks about economic impact when AI has offensive cyber capabilities - Anthropic conducted an internal review of AI usage at work - Review included a 132-person survey and 53 in-depth interviews - Review also analyzed internal Claude Code logs for systematic insights - Internal recap examines how AI in the workplace changes human roles

Key insights

Anthropic red team demonstrated an AI-enabled blockchain theft of $4.6M

Community reports state that an Anthropic red-team exercise used AI to attack a blockchain system and successfully exfiltrated $4.6 million. The research framed a central question about the economic consequences if AI systems obtain offensive cyber capabilities, contrasting this experiment with prior security work.

Anthropic performed a systematic internal review of AI usage and workforce effects

Anthropic carried out an internal review to understand how AI is used within the company and how work roles evolve. The review combined a 132-person questionnaire, 53 deep interviews, and an analysis of internal Claude Code logs to produce a comprehensive recap of AI’s role in employees’ daily tasks and workflows.

Source posts