Scan · Analyze · Deploy

Scans your models, datasets, and AI application code for malicious content. Never loads, imports, or runs what it inspects.

amara · run

$ amara run MustEr/gpt2-elite

[fetch] pytorch_model.bin 71.5 MB → quarantine

[format] pytorch-zip

CRITICAL PICKLE.DANGEROUS_GLOBAL

runpy._run_code → archive/data.pkl@2

payload: subprocess.run(['calc.exe'])

Findings: critical=1

Final Verdict: [MALICIOUS-BLOCK]

Why AMARA

One scanner for the whole AI supply chain

Model weights, datasets, and the LLM/Agent/MCP code around them, inspected without ever loading, importing, or running any of it.

Bytes only inspection, safe to point at a model you already suspect is malicious.

Every reference is treated as malicious until proven known safe.

Weights, datasets, and the LLM, agent, and MCP code around them, one report.

A policy engine, severity thresholds, and an exit code your pipeline can act on.

Explore

The Trojans within AMARA

Malware/backdoor scans, dataset poison detection, and LLM/Agent/MCP security scans, shown as real amara run sessions.

Real risks across Supply Chain, LLM, Agent, and MCP, How to Defend against each, mapped to the OWASP Top 10.

Every flag, every scanner, every finding severity, what it does, and why and when to reach for it.