ctf-ai-ml

SUSPICIOUS: internally consistent as a CTF attack reference, but it is an offensive security skill that equips an agent to perform AI/ML and LLM attack techniques against targets. Install trust is mostly normal package-manager use; the main concern is the exploit-oriented capability combined with Bash/Web access, not malware or covert exfiltration.

No evidence of stealthy system compromise (no backdoors, process control, credential theft, or covert exfiltration) is present in this fragment. However, the file is a set of active ML attack scripts (inversion, extraction, membership inference, and weight/LoRA manipulation) that can be directly abused, and it includes high-risk deserialization via torch.load of .pt artifacts, which can enable arbitrary code execution if model files are untrusted or tampered. Treat as potentially dangerous offensive tooling rather than a safe library; do not ship or execute with untrusted checkpoints.