Hacking Salesforce Sites With an LLM Agent

Summary: Reco's security research team built an autonomous LLM agent that takes only a Salesforce Experience Cloud URL and then maps the attack surface, analyzes endpoints, writes working exploits, and runs them without human guidance. Pointed at real sites, it found high-severity vulnerabilities at organizations with heavy security investment, including a broken access control flaw at a major cybersecurity vendor that returned full Contact records and billing addresses for any email address, and a blind SOQL injection at a Fortune 500 tech company.
What's compelling: At the vendor site, the agent autonomously pulled employee and partner PII and downloaded files marked "CONFIDENTIAL INFORMATION" through single unauthenticated HTTP requests, on a site where guest access to pages was disabled. The same capability that produced a responsibly disclosed report is available to threat groups like ShinyHunters, which has already scanned thousands of Salesforce Sites.

Read the original at The Hacker News ↗