If all the computing power in the world were brought together and the most powerful AI model were allowed to iterate and develop controlled nuclear fusion on its own, might this accelerate the pace at which humanity truly achieves a civilisational leap forward?
4
95
We have a new research named Ai2Shell.😉 – at Bundeena, New South Wales
🔬 New research: when an LLM is asked to "save this text verbatim," inference-framework parsers reinterpret embedded protocol markers as live tool-call control structure — silently injecting, rewriting, or dropping tool calls.
In our end-to-end lab, an agent with a shell tool received and executed arbitrary commands — triggered by a single sentence. The model recognized the injection, copied the payload anyway, and the server-side parser delivered the call. The agent never issued it.
We also showed tool_choice enforcement can be bypassed: server forces "only call tool X," parser delivers tool Y.
Full writeup: 2shell.ai/
Research by SEC4SI Team
@ProcessorLee · @VillanCh · @bestswngs · @chun_springX · @bbznop · @0chencc
3
26
SpringKill retweeted
🔬 New research: when an LLM is asked to "save this text verbatim," inference-framework parsers reinterpret embedded protocol markers as live tool-call control structure — silently injecting, rewriting, or dropping tool calls.
In our end-to-end lab, an agent with a shell tool received and executed arbitrary commands — triggered by a single sentence. The model recognized the injection, copied the payload anyway, and the server-side parser delivered the call. The agent never issued it.
We also showed tool_choice enforcement can be bypassed: server forces "only call tool X," parser delivers tool Y.
Full writeup: 2shell.ai/
Research by SEC4SI Team
@ProcessorLee · @VillanCh · @bestswngs · @chun_springX · @bbznop · @0chencc














