critical

CVE-2026-61732

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

Published 24 Sep 2026Modified 30 Sep 2026CWE 74

In the Trivy database since at least 24 Oct 2026, 07:11 UTCWhen it arrived before that build is not recorded.

Severity by source

GHSA10.0CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:H/A:H

Trivy's own label is CRITICAL. Rated critical by GHSA.

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.

Affected packages and fixed versions

Source	Package	Fixed in
pip::GitHub Security Advisory pip	decepticon, decepticon-core, decepticon-sdk	1.1.17

Show all 3 entries from 1 source

Source	Package	Fixed in
pip::GitHub Security Advisory pip	decepticon	1.1.17
pip::GitHub Security Advisory pip	decepticon-core	1.1.17
pip::GitHub Security Advisory pip	decepticon-sdk	1.1.17

References

What Attestkeep does with this

Inside your cluster, without you opening a spreadsheet.

KnowsThe scanner refreshes its vulnerability database every six hours.

FindsEvery admitted image is rescanned every 24 hours by default (scan.rescanInterval). When Trivy rates this vulnerability critical for the image's package source, it appears in the image's findings by name. Findings without a fix are left out unless includeUnfixed is on.

Tells youA critical_finding event goes to the channels subscribed to it: e-mail, Slack, Teams, a webhook, Jira, GitLab or GitHub.

RefusesWith a policy that blocks critical findings, a pod with this image is refused at admission, a rollback to it included. Pods already running are not evicted.

RecordsEach decision is recorded with the policy that ran and its hash, then sealed under a signed checkpoint.

Try Attestkeep

Read the docs