ResourcesWhite papers
White papers.
Long-form research on machine identity. Free to download with your work email.
- WP-01 · September 2026, revised October 2026 · 21 pages
The Agentic AI Threat Landscape
Agents do not fail like software. They fail like employees. Drawing on a verified record of 2026 incidents, this paper shows where identity governance decides what an attack is worth, where it does not, and what to do on Monday.
Contents- 01The short version
- 02The shift: agents as attack surface
- 03The 2026 incident record
- 04The trick and the prize
- 05The identity gap
- 06The regulatory vise
- 07The economic layer
- 08The control plane
- 09What to do on Monday
More from the research deskAll Field Notes →
- FN-14 · Oct 9, 2026Does the AI Kill Switch Work?A new benchmark tests whether a signal on a web page can halt a misbehaving AI agent. On average the best methods stopped fewer than half, and they stopped legitimate agents too. A revoked credential does not need the agent to agree.
- FN-13 · Oct 9, 2026How Many AI Agents Does Your Company Run?AI agents are multiplying inside enterprises faster than anyone can find them, name an owner, or retire them. Where they come from, what an uncounted agent costs, and six steps to get ahead of it.
- FN-12 · Sep 21, 2026Nobody Has Counted the MachinesPublished machine-to-human identity ratios disagree by an order of magnitude. We followed each one to its primary source, including our own, and found estimates, conference-floor samples, and a figure that is not on the page it cites.