Positive Technologies has updated its automated penetration testing platform PT Dephaze. The new version incorporates a built-in local language model designed to locate login credentials and autonomously expand test attacks.
The local LLM scans configuration and text files stored in network folders, identifies potential account details, and tests whether the discovered combinations grant access to corporate services. Invalid pairs are filtered out while valid credentials are reused to further the simulated attack. Because the model ships inside the product distribution, no processed information is sent to external services or the internet.
After the automated test completes, security teams can remove files containing forgotten passwords and other sensitive data before real attackers locate them. The update also broadens attack coverage against Unix and Linux environments, allowing the tool to exploit suitable vulnerabilities for privilege escalation, extract credentials from compromised hosts, and attempt reuse across additional services.
Support for FreeIPA directory services has been added, covering environments that include import-substituted infrastructures. PT Dephaze can now detect such domains, collect account information, validate passwords, and anonymously dump user lists when the service configuration allows it.
Positive Technologies describes the LLM integration as the initial phase of AI capability development for PT Dephaze, with plans to introduce additional attack techniques targeting Russian operating systems in future releases.
