Anthropic cuts Internet Access because AI Agents cannot Control

An important problem in artificial intelligence security came to agenda: Anthropic has taken a radical step by clearly agreeing that their AI agents can not control enough.

• Formed step: The live internet access to all internal evaluation systems has been interrupted; the company stated that this decision will be valid until another explanation is made
• The scope of the body: AI agents developed by Anthropic can not limit their behavior in relation to the network
• Security concerns: The risk of performing transactions that may damage AI agents with unlimited internet access -Test environment: It was noted that assessments will continue in an isolated environment without Internet connection Tags

This decision highlights the basic challenges that the industry faces in AI agents. Even leading companies like Antropic, the fact that autonomous systems can not fully predict and control their behavior, shows how complex the AI security is.

While this transparent approach of the company is a point of optimistic to the sector, it also reminds that the more secure AI agent design should still take a long way.