Decepticon is an open-source autonomous red team agent combining AI reasoning with security testing across 16 specialist agents, leveraging persistent tmux sessions and isolated Docker networks.
The tool was created to follow attack steps within predefined boundaries and rules. According to its creators, it has support for research, exploiting issues, gaining access, moving between devices, and remote control. However, these are claims made by the project and not proven effectiveness on real networks.
Operators can connect external programming tools to monitor progress, send instructions, and stop tests. The real work is performed within a dedicated test environment that isolates management services from the targets.
The system includes sixteen specialized AI agents for various domains, such as search, vulnerability research, and cloud systems. Each target is given a separate memory space to avoid information confusion, while design documents and rules of engagement are produced before each test.
The developers tested the tool in a specific evaluation test, where they achieved a high success rate. However, the tests themselves provided helpful hints that facilitated the research, so the results only apply to this controlled lab environment and not real-world business conditions.
As with any similar tool, teams should only use the software on systems for which they have explicit permission.
Top comments (0)