Last verified: October 30, 2026.
The short answer
McKinsey just had its internal chatbot hacked. It took two hours and twenty dollars for an offensive agent to expose over seven hundred thousand files and forty-six million chat logs through unauthenticated endpoints. Honestly, owning your software does not mean you own the security automatically. You have to configure it, monitor it, and treat private compute as a responsibility. That is exactly why buying it set up right matters so much.
The breach details
McKinsey released its internal chatbot named Lilli back in 2023. According to Consultancy.uk, the firm recently discovered a vulnerability that exposed over 728,000 private files and more than 46 million chat logs. The problem came down to twenty-two exposed endpoints that did not require authentication, including one with a SQL injection flaw. It took just two hours and twenty dollars in tokens for an offensive AI security agent to breach the system. McKinsey fixed all issues within hours of being notified by CodeWall. The firm stated its investigation found no evidence that client data or confidential information was accessed by unauthorized parties.
Why owning is not enough
This is the fourth piece in our private compute series, and it drives home a point I make often. You can own your software, you can host it on your own servers, but that does not mean the job is done. Private compute is a responsibility. When you deploy private AI infrastructure, you are taking on the burden of authentication protocols, endpoint security, and monitoring. If you leave endpoints open or skip proper configuration, you are handing attackers a key. I talk to business owners who think buying a system means they are safe. It does not. You have to secure it. You know what fixes that? Buying a setup that is already hardened, so you can focus on the math and the customers instead of playing sysadmin for an AI agent that costs twenty dollars to exploit.
What this means for you
If you are running a business today, here is what you need to watch for. You are the one deploying private AI infrastructure and configuring authentication protocols for internal tools. You cannot rely on the vendor to catch every misconfiguration. You need to monitor endpoint security for your AI agents just like you would for your payroll system.
- Audit your endpoints regularly. Make sure nothing is exposed without authentication.
- Verify SQL injection protections are active on any database connections.
- Treat private compute like cash flow. You own the asset, but you also own the risk. If something breaks, it hits your bottom line immediately.
- Consider external cybersecurity auditing services to catch vulnerabilities your internal team might miss.
The value of buying right
This is why I push transferability and systems so hard. When you buy a business, you want systems that are secure and reliable from day one. You do not want to inherit open doors. If you are building your own stack, you need to factor in the cost of security audits and ongoing maintenance. It is better to pay for a setup that includes professional auditing services upfront than to deal with the fallout after a breach. The math always favors prevention.
Frequently asked questions
What exactly was exposed in the McKinsey breach?
Over 728,000 private files and more than 46 million chat logs were exposed through unauthenticated endpoints.
How long did it take to breach McKinsey's AI system?
It took just two hours and twenty dollars in tokens for an offensive AI security agent to breach the system.
Did the breach compromise client data?
McKinsey reported that its investigation found no evidence that client data or confidential information was accessed by unauthorized parties.
What is private compute and why does it matter for business owners?
Private compute means you own and control your AI infrastructure rather than renting it. It gives you ownership of the data but also requires you to manage the security, configuration, and maintenance yourself.
How can small businesses secure their private AI deployments?
Secure deployments by enforcing authentication on all endpoints, monitoring for SQL injection flaws, and treating AI infrastructure with the same rigor as your financial systems. You may also need external cybersecurity auditing to catch what internal teams miss.
Sources
Get your Transferability Score
Free 14-question assessment. See where your business stands and what's holding it back from a clean exit.
Take the Free Score