Breaking

The Information's AI Deep Dive Podcast Examines How to Protect Advanced AI Model Weights

November 9, 2026 at 14:30 EST

Security
AI Agents

The Information has announced Episode 9 of its AI Deep Dive podcast, which focuses on how to keep advanced AI systems and their model weights secure.

Episode 9 · Podcast

Keeping advanced AI and its model weights secure

Lisa Thiergart, co-founder and CEO of Security Level Five, joins host Rocket Drew to discuss model exfiltration, the risk of rogue AI agents, and the technical challenges of building secure systems.

Three areas covered

Model exfiltration

Preventing unauthorized copying of model weights, the learned parameters that define how a model behaves.

Rogue AI agents

AI systems that take autonomous actions and deviate from their operators' intentions.

Technical challenges

How these risks can be engineered against, not only described.

Why security extends past the model

Model weights

Valuable assets that must be protected from theft.

→

Agents and tools

Access to data, tools, and networks widens the risk.

→

Engineered controls

Safeguards built into the systems around the model.

The central question

What would it take to keep advanced AI systems, and the model weights that define them, protected?

Commercial: stolen weights let a competitor or bad actor replicate a system.

Modification: a copied model can be altered without the original safeguards.

Safety: the same protections also bear on autonomous behavior.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.