"As a result, any extension can invoke a content script (which does not require any special permissions) and issue commands to the Claude extension. Claude in Chrome, it says, trusts the origin of the execution, which is claude.ai, and not the execution context, thus allowing any JavaScript running in the origin to issue privileged commands."
"This allows an attacker to create an extension with a declared content script and configured to run in the Main world, thus ensuring the script is executed as part of the page, and send a message to the Claude extension, which trusts the sender because it runs in claude.ai. Because a message handler in Claude in Chrome accepts and forwards arbitrary prompts, the attacker can perform remote prompt injection and control the AI agent's actions."
"While Claude enforces user confirmation for sensitive actions, as well as policies that prevent certain actions, and makes decisions based on certain inputs, LayerX discovered that the attacker's script could bypass these protections. The company was able to forge user approval by repeatedly sending a confirmation message and relied on Document Object Model (DOM) manipulation to dynamically modify UI elements and alter Claude's perception of the actions."
ClaudeBleed is a vulnerability in the Claude extension for Chrome that can let attackers take over the AI agent and steal information. The issue stems from lax permissions and trust decisions based on the origin of execution rather than the execution context. The extension allows interaction with any script running in the claude.ai origin without verifying the script owner. An attacker can create an extension with a content script that runs in the page’s main world, send messages to the Claude extension, and exploit a message handler that accepts and forwards arbitrary prompts. This enables remote prompt injection and control of agent actions. User confirmation and policy checks can be bypassed by repeatedly sending confirmation messages and manipulating the UI through DOM changes, while repeated triggering can reveal command execution effects.
#browser-extensions #ai-agent-security #prompt-injection #permission-and-trust-boundaries #web-ui-manipulation
Read at SecurityWeek
Unable to calculate read time
Collection
[
|
...
]