[US] AI agent breached Anthropic's internal network in 12 hours, exposing payroll records

In the US, the AI firm Anthropic's internal network was breached in under 12 hours by an AI agent from the cybersecurity startup Cathedral, during an officially authorised security test, BigGo Finance reports.

Among the data exposed by the breach were meeting transcripts, compensation and hiring information, and internal Slack messages in its sales, HR, and IT systems. 

Anthropic confirmed the intrusion, but stated that Claude user conversations, core source code, and model weights had not been leaked.

It has reportedly accused Cathedral, which was founded by former members of the U.S. Department of Government Efficiency, of exaggerating national security implications and violating testing rules. 

Despite this, the incident has led to concerns in Washington about the protection of frontier AI assets and third-party software risks.

Source: BigGo Finance

In the US, the AI firm Anthropic's internal network was breached in under 12 hours by an AI agent from the cybersecurity startup Cathedral, during an officially authorised security test, BigGo Finance reports.

Among the data exposed by the breach were meeting transcripts, compensation and hiring information, and internal Slack messages in its sales, HR, and IT systems. 

Anthropic confirmed the intrusion, but stated that Claude user conversations, core source code, and model weights had not been leaked.

It has reportedly accused Cathedral, which was founded by former members of the U.S. Department of Government Efficiency, of exaggerating national security implications and violating testing rules. 

Despite this, the incident has led to concerns in Washington about the protection of frontier AI assets and third-party software risks.

Source: BigGo Finance