OpenClaw has released version 2026.2.23 of its open-source personal AI assistant, focusing on enhanced security and advanced AI features. With over 215,000 stars on GitHub, the platform aims to address multiple vulnerabilities while introducing significant updates that cater to privacy-conscious users operating on macOS, Windows, and Linux.
One of the most notable enhancements in this update is the introduction of optional HTTP security headers, including Strict-Transport-Security. This feature supports direct HTTPS deployments, accompanied by comprehensive validation, tests, and documentation to help mitigate man-in-the-middle risks. Furthermore, developers have improved session maintenance through a feature called “openclaw sessions cleanup,” which introduces disk-budget controls and safer transcript handling. This aims to prevent data leaks and storage overflows.
Security Enhancements and Configuration Updates
A significant change in this release is the adjustment of the browser SSRF policy, which now defaults to “trusted-network” mode. Users on private networks will need to explicitly configure their settings, but they can migrate legacy configurations using the command “openclaw doctor –fix.” Additional fixes have been implemented to address both configuration and execution risks.
Sensitive dynamic keys, such as env.*, are now redacted in configuration snapshots, effectively preventing exposure while maintaining restore functionality. The system will block the execution of obfuscated commands without explicit approval, and the ACP client permissions will require trusted tool IDs with scoped read approvals to prevent unauthorized access to files.
OpenClaw’s security measures also include blocks against symlink escapes and XSS vulnerabilities in image galleries. With OTEL diagnostics, API keys are scrubbed from logs prior to export, ensuring enhanced protection against prompt injection and credential leaks in production environments.
AI Features and System Improvements
On the AI front, the update introduces first-class support for the Kilo Gateway with kilocode/anthropic/claude-opus-4.6 as the default integration. This includes improvements in authentication, onboarding, and cache handling. The Vercel AI Gateway has also been optimized to normalize shorthand Claude references. A new addition, the Moonshot “kimi” provider, enhances citation extraction capabilities in the tools/web_search feature.
The update expands media understanding by incorporating native Moonshot video support and refactoring execution to improve URL and header precedence. Agents now benefit from per-agent parameters to manage cache retention and bootstrap caching, minimizing prompt invalidations. Fixes also extend context pruning to Moonshot/Kimi and improve model resolution, making the system more reliable during high-demand scenarios.
The release of OpenClaw version 2026.2.23, which was tagged just hours ago by developer steipete, showcases contributions from numerous developers. This rapid evolution positions OpenClaw as a secure and versatile AI hub for messaging applications like WhatsApp and Telegram.
With targeted fixes for Telegram polling, adjustments to WhatsApp group policies, and accommodations for provider-specific quirks, this update ensures stable operations in an increasingly demanding ecosystem. For continuous updates on cybersecurity developments, users can follow OpenClaw on platforms such as Google News, LinkedIn, and X.
This latest release underscores OpenClaw’s commitment to providing robust security while integrating cutting-edge AI functionalities.
