cyfenced.com →

[ CVE-2026-61732 · Published 24 Sep 2026 ]

Decepticon: Role-boundary forgery via ChatML special-token literals in web crawl output composed into LLM context

CVSS 10.0 EPSS 1%Fix available

[ Why it matters ]Priority signals

  • Public proof-of-concept exploit exists (CISA SSVC)
  • Critical severity: CVSS 10.0

[ Description ]Official description

Decepticon is an autonomous hacking agent for red teams. Versions prior to 1.1.17 wrap web crawl results — the output of agent reconnaissance against target services — into LLM messages without neutralizing ChatML special-token literals. Under the BYOK (Bring Your Own Key) deployment model, users configure their own LLM credentials to any OpenAI-compatible endpoint. Most open-source and self-deployed model providers (vLLM, SGLang, Ollama, LM Studio, text-generation-webui, etc.) do not filter special-token literals from user content in their default configurations. Those literals are parsed into structural role-boundary token IDs, meaning an attacker string planted in a target web page forges a new operator turn the model treats as authoritative, bypassing Decepticon's agent guardrails and resulting in arbitrary command execution inside the Kali Linux sandbox. Version 1.1.17 patches the issue.

[ Fixes ]Fixes and patches

  • decepticon-core 1.1.17Fixed version · GitHub Advisories · pip package decepticon-core: upgrade to 1.1.17 or later (affected <= 1.1.16)
  • decepticon 1.1.17Fixed version · GitHub Advisories · pip package decepticon: upgrade to 1.1.17 or later (affected <= 1.1.16)
  • decepticon-sdk 1.1.17Fixed version · GitHub Advisories · pip package decepticon-sdk: upgrade to 1.1.17 or later (affected <= 1.1.16)

[ Packages ]Affected open-source packages

[ References ]References