Improper Neutralization of Special Elements in Output Used by a Downstream Component ('Injection') Affecting decepticon-core package, versions [,1.1.17)
Threat Intelligence
Snyk has a proof-of-concept or detailed explanation of how to exploit this vulnerability.
The probability is the direct output of the EPSS model, and conveys an overall sense of the threat of exploitation in the wild. The percentile measures the EPSS probability relative to all known EPSS scores. Note: This data is updated daily, relying on the latest available EPSS model version. Check out the EPSS documentation for more details.
Do your applications use this vulnerable package?
In a few clicks we can analyze your entire application and see what components are vulnerable in your application, and suggest you quick fixes.
Test your applications- Snyk IDSNYK-PYTHON-DECEPTICONCORE-20158964
- published27 Sept 2026
- disclosed24 Sept 2026
- creditUnknown
Introduced: 24 Sep 2026
NewCVE-2026-61732 (opens in a new tab)How to fix?
Upgrade decepticon-core to version 1.1.17 or higher.
Overview
decepticon-core is a Decepticon contract layer: pure types, protocols, plugin contracts, registry primitives
Affected versions of this package are vulnerable to Improper Neutralization of Special Elements in Output Used by a Downstream Component ('Injection') via UntrustedOutputMiddleware._format_envelope and PromptInjectionShieldMiddleware._wrap_untrusted, which emit attacker-controlled tool output verbatim without neutralizing chat-template special-token literals. A ChatML literal such as <|im_start|> planted in tool output (web crawl results, recon scan output, bash stdout) is passed to a self-hosted tokenizer that parses it into a structural role-delimiter token, forging an operator or system turn the model treats as authoritative. This bypasses the quarantine envelope and enables arbitrary command execution in the sandbox.
Note: This is only exploitable when Decepticon is configured with a BYOK OpenAI-compatible backend whose tokenizer preserves special-token IDs (vLLM, SGLang, TGI, and Ollama are confirmed vulnerable); hosted vendors such as OpenAI and Anthropic strip these tokens server-side.