Trust Me, I'm a Robot: Can we trust RPA with our most guarded secrets?

Robotic process automation (RPA) uses software bots that log into enterprise applications, so those bots must hold business credentials. The speakers chose Blue Prism, a platform named a market leader by industry analysts, and reverse engineered its .NET application server to see how it protects those secrets. In the design they examined, encryption keys sit on the application server while encrypted credentials live in a SQL Server database.

The talk walks through three attack paths. First, a server method that should have required authentication returned the master encryption key to an unauthenticated client, which was then used to decrypt stored credentials. Second, a secured method exposed a SQL Server system stored procedure that could run attacker-supplied commands, giving code execution on the database host and access to the encrypted credential store. Third, an insecure deserialization flaw in the WCF data contract serializer let an unauthenticated client on the domain inject a payload that was processed before the application's own permission checks, yielding administrator-level code execution on the application server.

The speakers argued that bots cannot use multi-factor authentication, and that compromising the application server or database exposes every credential and process the robots rely on. Their advice: follow vendor hardening guidance, apply defense in depth with network monitoring and endpoint detection, and keep credentials in external vaults rather than handing them directly to bots. They noted that the vendor had worked with them to release patches after disclosure.