Track vulnerabilities like this one.
Sign up free to get alerted when software you use is affected.
10.0
CVE-2026-61732: Decepticon can run attacker commands from web pages
CVE-2026-61732 · published 15 days ago
Summary
Versions of Decepticon before 1.1.17 insert raw web‑crawl results into the language‑model prompt without cleaning special tokens. If a malicious page’s text includes those tokens, the model treats them as trusted commands, allowing code execution inside the Kali Linux sandbox. Upgrade to version 1.1.17 or later, and ensure any custom model endpoint filters special tokens before processing.
What to do
- Update decepticon-core to version 1.1.17.
- Update decepticon to version 1.1.17.
- Update decepticon-sdk to version 1.1.17.
Affected software
| Ecosystem | Vendor | Product | Affected versions |
|---|---|---|---|
| – | bittersecurity | decepticon | < 1.1.17 |
| – | bittersecurity | decepticon-core | < 1.1.17 |
| – | bittersecurity | decepticon-sdk | < 1.1.17 |
| pip | – | decepticon-core |
<= 1.1.16 Fix: upgrade to 1.1.17
|
| pip | – | decepticon |
<= 1.1.16 Fix: upgrade to 1.1.17
|
| pip | – | decepticon-sdk |
<= 1.1.16 Fix: upgrade to 1.1.17
|
| PyPI | – | decepticon-core |
< 1.1.17 Fix: upgrade to 1.1.17
|
| PyPI | – | decepticon |
< 1.1.17 Fix: upgrade to 1.1.17
|
| PyPI | – | decepticon-sdk |
< 1.1.17 Fix: upgrade to 1.1.17
|
Original advisory text
Decepticon: Role-boundary forgery via ChatML special-token literals in web crawl output composed into LLM context
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.
References
- https://nvd.nist.gov/vuln/detail/CVE-2026-61732
- https://github.com/BitterSecurity/Decepticon/pull/715
- https://github.com/BitterSecurity/Decepticon/releases/tag/v1.1.17
- https://github.com/advisories/GHSA-g5f9-3xfg-p9mf
- https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/61xxx/CVE-2026-61732... Vendor Advisory
- https://github.com/BitterSecurity/Decepticon/commit/79ee2aaf22f4c36a5b1968f6ca3f...
- https://github.com/BitterSecurity/Decepticon/security/advisories/GHSA-g5f9-3xfg-...
- https://github.com/PurpleAILAB/Decepticon Product
- https://pypi.org/project/decepticon Product
- https://pypi.org/project/decepticon-core Product
- https://pypi.org/project/decepticon-sdk Product
Internet-facing
3 days
Internal
60 days
- Not known to be exploited
- Can be exploited automatically
- Gives an attacker full control
Type
CWE-74Injection
Timeline
Published24 Sep 2026
Updated9 Oct 2026
First seen24 Sep 2026
Sources
CVE-2026-61732 · NVD
CVE-2026-61732 · MITRE
GHSA-g5f9-3xfg-p9mf · GHSA
CVE-2026-61732 · OSV
GHSA-g5f9-3xfg-p9mf · OSV
PYSEC-2026-4034 · OSV
Track software like this
Free during beta