CVE-2025-64320
📋 TL;DR
This vulnerability allows attackers to inject malicious code into Salesforce Agentforce Vibes Extension's LLM prompting system. When exploited, it could lead to unauthorized actions or data exposure. All users of Salesforce Agentforce Vibes Extension versions before 3.2.0 are affected.
💻 Affected Systems
- Salesforce Agentforce Vibes Extension
⚠️ Risk & Real-World Impact
Worst Case
Complete compromise of the Salesforce instance, allowing attackers to execute arbitrary code, access sensitive customer data, and potentially pivot to other systems.
Likely Case
Limited code execution within the Salesforce environment, potentially allowing data exfiltration or manipulation of business processes.
If Mitigated
With proper input validation and output encoding, the vulnerability would be prevented from being exploited.
🎯 Exploit Status
Exploitation requires understanding of LLM prompt injection techniques and access to the vulnerable interface.
🛠️ Fix & Mitigation
✅ Official Fix
Patch Version: 3.2.0
Vendor Advisory: https://help.salesforce.com/s/articleView?id=005228032&type=1
Restart Required: No
Instructions:
1. Log into Salesforce Admin console. 2. Navigate to AppExchange packages. 3. Locate Agentforce Vibes Extension. 4. Upgrade to version 3.2.0 or later. 5. Verify the update completed successfully.
🔧 Temporary Workarounds
Disable LLM Features
allTemporarily disable the LLM prompting functionality in Agentforce Vibes Extension
Restrict User Access
allLimit access to the Agentforce Vibes Extension to only essential users
🧯 If You Can't Patch
- Implement strict input validation and sanitization for all LLM prompt inputs
- Monitor logs for unusual LLM prompt patterns or unexpected code execution attempts
🔍 How to Verify
Check if Vulnerable:
Check the installed version of Agentforce Vibes Extension in Salesforce Setup > Installed PackagesCheck Version:
Not applicable - check via Salesforce Admin interfaceVerify Fix Applied:
Confirm version is 3.2.0 or higher in Salesforce Setup > Installed Packages📡 Detection & Monitoring
Log Indicators:
- Unusual LLM prompt patterns
- Unexpected code execution in Salesforce logs
- Multiple failed prompt attempts
Network Indicators:
- Unusual API calls to Salesforce LLM endpoints
- Suspicious data exfiltration patterns
SIEM Query:
source="salesforce" AND (event_type="llm_prompt" OR event_type="code_execution") AND status="unusual"