Salesforce Agentforce Vulnerabilities Enable 0-Click Data Theft

• •Three vulnerabilities in Salesforce Agentforce enable zero-click data exfiltration.
• •Attackers can exploit public Web-to-Lead forms to inject malicious instructions.
• •Salesforce has remediated the issues in collaboration with Zenity Labs.

Zenity Labs disclosed three vulnerabilities in Salesforce's Agentforce, collectively termed SalesBleed, which allow attackers to exfiltrate sensitive CRM data without user interaction. The attack exploits a public Web-to-Lead form to inject malicious instructions that remain dormant until a user queries the agent. Once triggered, the agent processes the poisoned lead and executes hidden commands, leading to unauthorized data transmission to attacker-controlled servers. Salesforce has collaborated with Zenity to address these vulnerabilities by enhancing its Trusted URLs mechanism. The vulnerabilities highlight the challenges of securing AI agents and the need for continuous monitoring. The attack method involves zero-click data exfiltration and AI agent impersonation for phishing. Currently, the vulnerabilities have been patched, and the attack vectors are no longer effective.