SMRTR summary
Claude Code (Opus 5) can be hijacked to execute malicious code simply by summarizing a poisoned website, chaining a redirect, malicious ZIP, and Python module shadowing—bypassing safety guardrails 60–80% of the time. Anthropic says it's working as designed.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original articleThis archive is built from SMRTR newsletter stories. Subscribe for hand-picked stories without the extra noise.
Anthropic released Claude Opus 4.5, positioning it as the world's best coding AI model, now with desktop Claude Code support.
Anthropic's new Claude Code auto mode lets AI act independently while blocking risky operations like file deletion and data transmission.
Claude Code, Anthropic's new tool, uses advanced language models to execute coding tasks, leveraging Claude 3.7 and 3.5, supporting AWS Bedrock and Google Vertex AI, and offering...
Anthropic's Claude Code, a powerful AI coding assistant, offers advanced features like subagents, custom commands, and context management. It excels at code generation and...
Anthropic launches Claude's computer control features with safeguards, but warns protections aren't perfect and recommends avoiding sensitive data.
Anthropic's new Claude Code 'auto mode' automatically approves safe actions while blocking risky ones, reducing interruptions during development.