OpenClaw, the open-source autonomous AI assistant that has gained widespread adoption in early 2026, launched model v2026.2.17 on February 17, 2026, introducing assist for Anthropic’s newest Claude Sonnet 4.6 mannequin.
The discharge comes amid rising safety issues after researchers documented the primary in-the-wild credential theft concentrating on OpenClaw configuration information by infostealer malware.
New Anthropic Mannequin Integration
The v2026.2.17 replace expands OpenClaw’s model-agnostic structure to assist Anthropic’s newly launched Claude Sonnet 4.6 mannequin, which customers had beforehand encountered errors when making an attempt to make use of.
OpenClaw already supported varied Anthropic Claude fashions together with Opus and Sonnet variants by means of API key authentication, with options like prolonged immediate caching and entry to Anthropic’s beta 1-million-token context window.
New Options
The framework permits customers to configure Anthropic fashions through CLI setup or direct config file modification, with choices for various cache retention durations and enhanced context capabilities.
Safety researchers at Hudson Rock reported the primary documented case of infostealer malware efficiently exfiltrating OpenClaw configuration information from a sufferer’s system.
The malware captured delicate information together with openclaw.json containing gateway authentication tokens and consumer credentials, system.json with non-public cryptographic keys used for system pairing and signing, and reminiscence information storing private context like each day actions and personal messages.
The stolen gateway token might allow attackers to remotely connect with the sufferer’s native OpenClaw occasion if uncovered or impersonate the consumer in authenticated requests.
This incident marks a major shift within the risk panorama as AI brokers grow to be extra built-in into skilled workflows.
Safety specialists warn that infostealer builders will probably create devoted modules particularly designed to goal OpenClaw information, just like present modules for Chrome or Telegram.
The assault was described as a “grab-bag” operation the place the infostealer used broad file-harvesting routines that unintentionally captured the entire operational atmosphere of the sufferer’s AI agent.
OpenClaw runs domestically on consumer machines with full system entry capabilities together with file operations and shell instructions, making correct safety configuration crucial.
Comply with us on Google Information, LinkedIn, and X to Get Immediate Updates and Set GBH as a Most well-liked Supply in Google
