genai / news / / TNW
Hugging Face hosts more than a million machine learning models used by virtually every AI company on the planet.
Hugging Face hosts over 51,700 models with 352,000 unsafe or suspicious issues identified by Protect AI.
KEY POINTS
- Attackers use the nullifAI technique, combining 7z compression with malicious Python pickle code to evade detection.
- ClawHub's skill registry was infiltrated by 'ClawHavoc,' a coordinated campaign planting 335 malicious AI agent skills.
- 36% of all AI agent skills examined by Snyk contain security flaws, with 20% classified as malicious.
- Malicious payloads in AI models and skills execute immediately on load, often before human review occurs.
COMPANIES
Summarized by Newsio from TNW. How we summarize →