Description

Recent research by an independent AI safety and security research lab, TheWeatherReport.ai, by MIT research affiliate and former Google senior manager of cloud AI safety and security, showed that AI code security gains are not keeping pace with functionality gains between generations of frontier models. He talks through the implications of these findings, how IT organizations can protect themselves, and his outlook for how AI safety and security will continue to develop.

Featuring: Ilya Kabanov, founder, TheWeatherReport.ai

In today’s episode, we’ll cover…

• What surprised Kabanov about changes in AI code security
• The pros and cons of prompts, deterministic rules and "human-in-the-loop" for AI code security
• AI safety and security in the wake of the OpenAI / Hugging Face incident

And more!

References:

• OpenAI models escape containment, hack Hugging Face
• AgenticOps for infrastructure faces a deterministic dilemma
• Cisco agentic AI security push faces enterprise trust gap
• Capability without security: measuring the functionality-security gap in AI-generated code

To learn more about software development and platform engineering, check out Search IT Operations.

To watch clips from our podcast, subscribe to our YouTube channel, @EyeOnTech.