5 Signs an AI-Suggested Package Might Be Malicious
Not every real package is a safe one. Five red flags that an AI-suggested PyPI package could be a slopsquat — and how mirago spots them.
A package existing on PyPI doesn't make it safe — an attacker can register a fake-sounding name and fill it with malware. Here are five signals that an AI-suggested package deserves a second look.
1. It was created very recently
Legitimate, widely-used packages have history. A package published days ago that an AI just told you to install is worth scrutinising. mirago flags packages created less than 30 days ago as one risk factor.
2. Almost nobody downloads it
Real, useful libraries accumulate downloads. Near-zero download counts on a package you're told is essential is a red flag — mirago treats fewer than ~100 downloads/month as the second factor.
3. It's not already in your project
If your lockfile doesn't list it and you don't recognise it, you're trusting the AI alone. mirago treats anything already in your lockfile as trusted and focuses suspicion on the unknowns.
4. The name is suspiciously on-pattern
Slopsquats are designed to look right — async_json_fast, secure_requests2. Plausibility is the point. Cross-check the name against the project's real homepage and repository.
5. The signals stack up together
Any one signal alone is weak. mirago only raises a warning when a package is recently created and barely downloaded — both at once — so it stays quiet on harmless new libraries and speaks up on the genuinely sketchy ones.
```bash
mirago check .
```
That conservative, multi-signal approach (how it works) is deliberate: a tool that cries wolf gets uninstalled. Run the check before you install, every time.