
AI Agents, Real Authority, and Old Vulnerabilities
What happens when an ideation workshop produces working prototypes before the first day is over? Richard shares how collaboration between IBM and Microsoft helped shape IBM Imagine Enterprise, and why coordinating agents as a workforce brings harder questions than building them individually. He is also candid about the work still ahead.
Cyrus explores why industrial systems need remote connections, and why securing equipment with decades of service behind it takes more thought than telling someone to install an update. The discussion covers controlled supplier access and the importance of alarms when digital interference has physical consequences.
David examines research describing agents hunting exposed GitHub credentials and following through into attacks. Deleting the evidence from a repository offers little comfort if the credential still works. Help desk impersonation and certificate services bring the discussion back to older infrastructure, while a brief mention of Microsoft’s sleeper-agent research adds another reason to keep controls outside the model.
Links
- Microsoft Research, Lazarus and dormant-neuron backdoors in AI agents
- IBM and Microsoft, Imagine Enterprise announcement
Socials
- X/Twitter: link unavailable during saving
- Bluesky
- Facebook: link unavailable during saving
- Threads
Music
Null Invocation, Monochrome Pulse Music link
More episodes from "Cloudy with a Chance of Insights | A Microsoft Cloud (Azure, M365) Show for Architects & Engineers"



Don't miss an episode of “Cloudy with a Chance of Insights | A Microsoft Cloud (Azure, M365) Show for Architects & Engineers” and subscribe to it in the GetPodcast app.








