OpenClaw 2.0 pours glitter on slow-burning security dumpster fire
OpenClaw 2.0, an update to the popular AI agent harness, focuses on improved usability with a redesigned interface and simplified installation. However, security concerns remain largely unaddressed, despite new features like protected credentials and a contributor-controlled sandbox. The update prioritizes accessibility and ease of use over robust security measures, raising concerns about potential misuse and data exposure.
OpenClaw has unveiled version 2.0 of its AI agent harness, prioritizing user experience improvements while largely neglecting security enhancements. The update aims to simplify installation and provide a more familiar interface, resembling popular AI services like ChatGPT and Gemini.
“This update touches every part of OpenClaw,” Foundation community manager Hannes Rudolph explained. “We started by simplifying installation and rebuilding the browser app as a first-class experience, but doing that properly meant carrying the cleanup through the rest of OpenClaw until it became OpenClaw 2.0.”
Key features include a redesigned browser app with a conversational interface and simplified installation, allowing users to quickly engage with their AI agents. A new shared cloud sessions feature enables collaborative use of OpenClaw, maintaining context across multiple users.
Despite these additions, security concerns persist. OpenClaw has previously been criticized for its potential to expose user data, as demonstrated by tests where it shared private information and even hacked a gym’s waiting list. The update’s security improvements are minimal, with the patch notes highlighting that shared session controls are not tenant isolation and that secret store values are not encrypted at rest.
Furthermore, a new contributor-controlled sandbox, intended to isolate untrusted code, is disabled by default. The update essentially prioritizes ease of use and accessibility over comprehensive security, raising concerns about the potential for misuse and data exposure when granting widespread access to this powerful tool.