Skip to content

BRK466 Breakout Thu 19 Nov 11:30 PST

Securing edge AI and operational technology (OT) workloads

Deepak Manohar, Dona Mukherjee, Fred Reiss

Summary

From the recording

Speakers from Microsoft and IBM Research covered securing industrial assets, models and data across Azure Local, Foundry Local and Azure IoT. Announced: IoT certificate management is GA; AMD confidential VMs are GA in Azure Local, with Intel in preview; confidential compute for data services such as SQL is in public preview; Foundry Local guardrails start as an initial preview. Bring Your Own Key with AKV was stated as GA. Dell, HP and Lenovo were named as OEM partners, and Forescout joined the Azure IoT partner programme. Captions are auto-generated, and no dates were given for planned items.

Session description

Microsoft's catalogue description

As manufacturers move AI models and sensitive workloads to the edge, cybersecurity risk crosses the IT/OT boundary. Learn how to protect models and operational data, enforce Zero Trust controls, and build resilient architectures across segmented OT networks without disrupting production or safety. Real-world examples show how teams move edge AI from pilot to production with clear governance and recovery practices.

Demo timestamps

Microsoft's player does not support links to a specific time, so open the recording and skip to the time shown.

  1. Skip to 0:10:01

    Device registry firmware analysis and update

    In Azure Device Registry, firmware 1.4.0 shows 18 known-exploited and 354 critical vulnerabilities, weak signatures and expired certificates. An update job moves it to firmware 1.5.0 with no critical or high vulnerabilities left.

  2. Skip to 0:21:38

    Protecting the model with a confidential VM

    An app on a confidential VM deploys an encrypted model decrypted only after attestation, with Sigstore signing verifying its origin. IBM's Granite Switch packs several adapters into one artefact.

  3. Skip to 0:29:41

    Foundry Local guardrails block prompt injection

    Operators at a fictional recycling company query sensors in a chat app. A prompt injection extracts confidential readings; with guardrails on, the same attack is blocked by policy.

  4. Skip to 0:34:24

    End-to-end factory scenario with trust chain

    An alarm on pump 104 leads the AI to recommend pausing it based on verified readings. The demo shows the trust chain behind the answer, including protected ingestion credentials whose secret values stay hidden.

Watch on-demand (opens in a new tab)

← All sessions