CVE Database

Improper Neutralization of Special Elements in Output Used by a Downstream Component in the colour control escape code handler in kitty from 0.47.3 before 0.49.0 allows a program writing to the terminal to execute an arbitrary command in the user's shell, because color_control() in kitty/window.py answers a query for an unrecognised field name by placing that field name into the reply, and write_escape_code_to_child() in kitty/screen.c then writes the reply to the pseudoterminal master, where it is not distinguishable from input typed by the user, without neutralising it for the shell that reads it. The payload is reduced to printable ASCII before the field name is echoed, which is the restriction introduced in 0.47.3 as the fix for CVE-2026-54057, and the record and field separators ; and = are consumed as delimiters, but every other printable character survives, which is sufficient to compose a shell command. A newline is available from handle_remote_ssh() in kitty/window.py, which writes the bytes yielded by get_ssh_data() in kittens/ssh/utils.py, the first of which begin with a newline, to the pseudoterminal master before any credential carried in the request is checked. The reply is framed as an OSC sequence carrying the escape code number, the field name, and the literal value ?. This results in execution of an attacker-chosen command with the privileges of the user running the terminal.

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.