Claws are now a new layer on top of LLM agents
twitter.com
February 21, 2026
2 min read
Summary
Andrej Karpathy purchased a new Mac mini to experiment with OpenClaw. He expressed concerns about privacy and security risks associated with OpenClaw, citing reports of vulnerabilities and exposed instances.
Key Takeaways
- Andrej Karpathy purchased a new Mac mini to experiment with AI tools called "Claws," which are gaining popularity.
- He expressed concerns about security vulnerabilities associated with running OpenClaw, including exposed instances and remote code execution risks.
- Karpathy noted that smaller Claw projects, like NanoClaw, are emerging, featuring manageable code sizes and innovative configurability through skills rather than traditional config files.
- He highlighted the appeal of local setups for Claws, as they facilitate easy connections to home automation devices and provide a tangible digital assistant experience.
Community Sentiment
MixedPositives
- Implementing a one-time password requirement for agents can significantly enhance safety, ensuring user awareness and control over potentially harmful actions.
- The concept of 'claws' could lead to more intuitive user-agent interactions, potentially simplifying how users navigate complex tasks with LLMs.
Concerns
- The architecture of agentic workflows has a critical security flaw, as unrestricted access can lead to inevitable arbitrary code execution risks.
- Current implementations of agents may overlook significant vulnerabilities, such as context pollution, which can compromise sensitive data security.
Source
twitter.com
Published
February 21, 2026
Reading Time
2 minutes
Relevance Score
65/100
🔥🔥🔥🔥🔥
Why It Matters
This page is optimized for focused reading: quick context up top, a clean summary block, and a direct path to the original source when you want the full story.