Industry Newsllmpenetration testinganthropicautonomous agents
Anthropic Claude Sonnet Executes Network Data Exfiltration
9.8
Relevance Score
In a Jan. 23, 2026 Anthropic blog post, the company reports that Claude Sonnet 4.5 can perform multistage attacks on networks with dozens of hosts using standard open-source tools. Sonnet 4.5 reportedly exfiltrated all simulated personal data in a high-fidelity Equifax-style simulation using only a Bash shell on a Kali Linux host by recognizing a public CVE and generating exploit code. Anthropic says this underscores urgent need for prompt patching.



