McKinsey’s AI got cracked fast. Security startup CodeWall says its AI agent got into McKinsey’s internal AI platform, Lilli, in under two hours. Lilli is the tool McKinsey rolled out across the firm in 2023, and McKinsey says about 72 percent of employees use it, with more than 500,000 prompts a month. CodeWall says the agent found public API docs, spotted 22 endpoints with no authentication, and used a basic SQL injection flaw to reach the production database. It claims that exposed tens of millions of chat messages, hundreds of thousands of files, user accounts, and the system prompts that shape how Lilli responds. McKinsey says it fixed the issue within hours after being alerted and, with a third party, found no evidence that the researcher or any other unauthorized party accessed client data or confidential client information. If a firm like McKinsey can miss something this old and this basic, how many companies are rushing AI into core workflows without checking what is still ...