CVE-2025-54795: high-severity vulnerability in anthropics claude-code
Claude Code echo command allowed bypass of user approval prompt for command execution
Published
E-mail alert when it lands in KEV, gets a public PoC or its score goes up.
No sign of exploitation. No public exploitation artifact known so far.
Claude Code has a flaw that lets attackers run commands without getting permission from the user. An attacker needs to insert malicious content into the chat, and the tool will execute their command without asking for approval first.
CWE-78 command injection vulnerability in Claude Code versions <1.0.20 allows bypass of user confirmation prompts through a command parsing error. Attack vector requires injecting untrusted content into the context window; impact is arbitrary command execution with user privileges without explicit approval.
In the same product, most dangerous first.