BlockBeats news: On February 26, Peter Steinberger, founder of OpenClaw, stated in a recent interview that OpenClaw currently offers the most comprehensive privacy solution, but security is a different matter. All data is stored locally, with nothing uploaded to the cloud; users have full control over access permissions and memory data, ensuring privacy.
But the security risk isn’t about being breached—it’s about losing control. Peter says the security of AI agents largely depends on the capabilities of the underlying model. Prompt injection attacks against large models are not impossible, but they are not as easy as many people assume. OpenClaw has now brought on dedicated security experts whose core focus is helping users operate safely in new scenarios.
Security concerns surrounding OpenClaw remain troubling. Data from February revealed 341 malicious plugins in the skills marketplace, with a contamination rate of 11.3%, creating significant supply chain risks. However, under the conditions of using the latest models and proper configurations, AI agents demonstrate stronger self-defense capabilities than commonly perceived.