Calif

Security work on and with AI

Calif's work includes attacking AI models and agents, evaluating offensive capabilities, and using AI in vulnerability research. It also provides infrastructure and product security engineering. Those activities have different targets: testing an agent's behavior is distinct from using that agent to investigate a bug in other software. [1]

The PHP2Shell investigation

Calif contributed to the October 2026 PHP2Shell disclosure, in which researchers investigated flaws with frontier models and coordinated with PHP maintainers and other security teams. Its published account explains the investigation rather than simply asserting that an automated scanner found a bug. [2][3]

Discovery, validation and patching

The coordinated response connected researchers, model developers and the people maintaining affected software. A model-assisted finding still needed validation, an assessment of affected configurations and a patch. Calif's role in that specific investigation is stronger event evidence than a customer logo or a general statement that it works with AI labs. [2][3]

The research demonstrates particular attack paths. It should not be read as a count of exploitable public websites or as proof that every affected installation exposed the same path. That deployment context matters when interpreting both the security impact and the contribution of AI. [2]