Update Claude setups now: Anthropic cuts live web browsing after failing safety checks

Anthropic has restricted the live internet browsing capabilities of its Claude AI model following failures in internal safety evaluations. This decision directly affects how system administrators and security teams deploy the technology within their organizations. The company implemented these limits to address specific safety concerns identified during testing. According to Security Affairs, the restriction was highlighted in their weekly newsletter Round 599, signaling a shift in how the tool is managed.

In plain English

Claude is an artificial intelligence model developed by Anthropic that can process information and generate responses. One of its features allows it to access the live internet to retrieve current data. This capability enables the model to provide up-to-date answers rather than relying solely on its training data. However, accessing the live web introduces risks. The AI might retrieve harmful content or be manipulated into performing unintended actions. Anthropic determined that the current safety measures were insufficient to prevent these risks during live browsing.

The background

Security evaluations are critical tests that vendors run to ensure their products behave as intended. These tests check for vulnerabilities, misuse potential, and safety guardrails. In this case, Claude failed these specific evaluations related to live internet access. The failures indicated that the model could not reliably distinguish between safe and unsafe web content or actions. Consequently, Anthropic decided to limit this feature to mitigate the risk. This move reflects a broader industry trend of tightening controls on AI capabilities that interact with external systems.

What changes now

Organizations using Claude must adjust their deployment strategies to account for the restricted internet access. System administrators need to review their current configurations and usage patterns. If workflows relied on Claude’s live browsing for real-time data retrieval, those processes will need alternative solutions. The restriction applies to the live web browsing feature specifically. Other core functionalities of the model remain available, but the scope of what the AI can do autonomously on the internet has been reduced. Teams should verify their vendor contracts and service level agreements to understand the full impact.

What to do and how to stay safe: Anthropic

• Review current Claude deployment configurations to identify any reliance on live internet browsing features.
• Assess workflows that depend on real-time data retrieval and plan alternative methods for information gathering.
• Monitor vendor communications for further updates on safety measures and feature availability.
• Educate users on the limitations of the AI model to prevent misuse or incorrect expectations.

Step-by-step guide: Stop AI Chatbots Leaking Sensitive Business Data

General security guidance from the Vector Update newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

Why did Anthropic restrict Claude’s internet access?

Anthropic restricted the feature because Claude failed internal safety evaluations related to live web browsing.

Does this affect all Claude features?

The restriction specifically targets live internet browsing; other core functionalities remain available.

How can organizations adapt to this change?

Organizations should review their deployment strategies and find alternatives for workflows requiring real-time data.