Securing AI: Threat Modeling & Detection

23 snips

May 27, 2025

Joining the discussion are Jackie Bow, a technical lead at Anthropic focused on threat detection, and Kane Narraway, head of enterprise security at Canva specializing in zero trust. They debate whether AI is a security threat or an innovative ally. Jackie reveals how to harness AI, even its quirks, for effective threat detection, while Kane emphasizes the need for robust threat modeling amid new risks. They also tackle the complexities of AI in production and whether AI tools enhance or jeopardize security skills, all served with a dash of humor.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

ADVICE

Control AI in Detection

Use large language models with visible prompts to control and tweak AI outputs.
Apply "best of end" techniques by having multiple model runs and selecting the best response.

INSIGHT

Embrace Controlled Hallucinations

Allow AI models to hallucinate creatively within controlled boundaries.
Creativity from AI hallucinations can lead to novel investigative ideas not previously considered.

ADVICE

Threat Model AI by Access & Integration

Start AI threat modeling by focusing on access points and integrations.
Prioritize securing AI connections to data stores and third-party systems to reduce risks.

Get the Snipd Podcast app to discover more snips from this episode

Get the app