BRK212 Breakout Wed 18 Nov 15:00 PST
Secure the Follow-Through: Trusting Autopilot at Scale
Eric Sachs, Erica Toelle
Summary
From the recording
Eric Sachs set out four principles for trusted autonomous agents: unique identity, bounded access and governance, continuous protection, and transparency. Microsoft Autopilot has its own email address, Teams handle and OneDrive files, and its core agent runs in an isolated container without credentials. Sachs said existing human security practice gets you 80 to 90% of the way, and that 9 times out of 10 an agent incident can be treated like a vendor incident. Erica Toelle said Autopilot will be in limited preview for quite a while. No dates, pricing, licensing or regional availability were given; product names may be mis-transcribed.
Session description
Microsoft's catalogue description
Autopilot keeps work moving, taking action across apps, data, and devices without constant direction. But how do you secure work that continues when you're away? Learn how to protect Autopilot's access and actions, safeguard sensitive data even when AI is misled, and contain activity across computers, apps, and networks. Because when AI works independently, security can't be an afterthought.
Demo timestamps
Microsoft's player does not support links to a specific time, so open the recording and skip to the time shown.
- Skip to 0:19:07
Agent identity and accountability in Microsoft Entra
Shown from an administrator perspective: identifying an agent and its accountable owner. Autopilot identities appear in Microsoft Entra and behave like human accounts, so permissions, governance and investigations apply consistently.
- Skip to 0:26:27
Governance, access packages and continuous protection with Defender
Access packages bundle SharePoint and Microsoft 365 group permissions with automatic expiration and approval policies. Defender and Agent 365 are used to investigate an incident and block one Autopilot instance without disabling all agents.