A critical vulnerability in BitterSecurity Decepticon allows unauthenticated attackers to bypass agent guardrails and execute arbitrary commands within the target sandbox environment.

The application fails to sanitize web crawl results before processing them as LLM messages, allowing ChatML special-token injection. This flaw permits an unauthenticated attacker to manipulate the model into accepting forged operator instructions, resulting in unauthorized command execution.

This vulnerability carries a CVSS score of 10.0, reflecting its critical nature and ease of exploitation. Successful exploitation grants an attacker full control over the Decepticon autonomous agent, which could lead to the compromise of internal network reconnaissance data, unauthorized access to sensitive target systems, and potential lateral movement from the Kali Linux sandbox into the broader corporate environment.

Immediate Action: Upgrade all components of the Decepticon suite, including decepticon, decepticon-core, and decepticon-sdk, to version 1.1.17 or later immediately.

Proactive Monitoring: Review LLM interaction logs for unexpected structural role-boundary tokens or anomalous command patterns that deviate from standard reconnaissance behavior.

Compensating Controls: Implement input validation at the web proxy level to strip ChatML special-token sequences from incoming traffic before it reaches the LLM ingestion pipeline.

Public Exploit Available: No

Given the critical severity and the potential for complete system compromise, organizations currently utilizing Decepticon must prioritize this update above all other maintenance tasks. The ability for an unauthenticated attacker to inject authoritative commands into an autonomous security agent represents a significant threat that necessitates immediate patching and a thorough review of existing agent deployment configurations.