Salesforce Agentforce Vulnerabilities Enable 0-Click Data Theft and Phishing

• •SalesBleed vulnerabilities allow 0-click data theft and phishing via Salesforce Agentforce.
• •Attackers can exploit public Web-to-Lead forms to exfiltrate sensitive CRM data.
• •Salesforce has patched the vulnerabilities but highlights ongoing risks with AI agent security.

Zenity Labs has uncovered multiple vulnerabilities in Salesforce's Agentforce, collectively termed 'SalesBleed.' These flaws allow attackers to exploit public Web-to-Lead forms to execute 0-click data exfiltration and hijack trusted AI agents for phishing. Specifically, two vulnerabilities enable silent extraction of sensitive CRM data without user interaction, while a third allows phishing messages to be sent via Slack under the agent's identity. Salesforce has acknowledged these issues and collaborated with Zenity to implement fixes. The vulnerabilities highlight significant challenges in securing AI agents and maintaining control over their access. Current status indicates that the vulnerabilities have been patched, but the potential for exploitation remains a concern. Organizations using Salesforce Agentforce should review their security measures to mitigate risks associated with these vulnerabilities.