Claude helped three security researchers go from a vulnerable image upload to OpenAI’s internal code repository in under 72 hours. The breach still required skilled human guidance, but it shows how AI can compress exploit development while coding-agent integrations expand what a compromised account can reach. https://cryptoslate.com/anthropics-claude-helped-3-researchers-breach-openai-in-under-72-hours/