Research

Longer analysis of AI risk, controls and regulation, written for the people who have to act on it: security, risk, compliance and legal teams.

All downloads

Five white papers on the AI security incidents of autumn 2026, each with dated sources, a control map and a 90-day plan or design checklist, plus our free guides. All free, no sign-up.

White paper 01 | AI governance

The AI Agent Nobody Owns

AI agents now log in, run commands and fill in forms, often with real access and no owner. How to govern every agent as a privileged identity.

PDF, 9 pages. v1.1, 25 Oct 2026.

White paper 02 | AI security

Your AI Stack Is the Supply Chain

Gateways, coding agents, plugins, MCP servers and open models were all attacked in autumn 2026. A six-layer map and the controls most vendor reviews miss.

PDF, 9 pages. v1.1, 25 Oct 2026.

White paper 03 | AI security

Indirect Prompt Injection in the Wild

Injection now turns up on web pages, in HTTP headers and in phishing emails. Six patterns, why filters fail, and a design checklist for AI features.

PDF, 9 pages. v1.1, 25 Oct 2026.

White paper 04 | AI governance

When the AI Vendor Is the Incident

Often the AI risk sits with the provider, not your staff. Who is accountable under the EU AI Act and the GDPR, plus a vendor due-diligence framework.

PDF, 8 pages. v1.1, 25 Oct 2026.

White paper 05 | AI threats

The Attacker Has the Same Models

AI shrinks the time between a weakness existing and being used. What that changes for patching, detection, AI keys and deepfake fraud.

PDF, 9 pages. v1.1, 25 Oct 2026.

Free guide | AI governance

AI Governance, the Short Version

A plain-English map of the whole field, built on the AIGP syllabus: 58 topics in 79 short lessons, a few sentences each. Read it in an evening.

PDF, 26 pages. Current as of 25 Oct 2026. Unofficial study material, not affiliated with the IAPP.

Our standards

  • Method first. Every report explains how the work was done and where its limits are.
  • Sourced and dated. Claims link to their sources and say when they were true.
  • Independent. No sponsor or client ever changes a finding.
  • Corrected in public. If we get something wrong, we fix it and note the change on the report.

Versions and corrections

All five white papers are version 1.1, published 25 October 2026. Every fact in them was checked against the archived text of its source (or the live page, where noted) on that date. If you spot an error, email danilo@peritumai.com. We fix mistakes in public and list each change here.

Version 1.1 (25 October 2026). After an independent review against every source, we corrected wording that overstated what sources said, labelled two sponsored sources as sponsored, separated four related agent-tool flaws from prompt-injection cases in paper 03, corrected one description of a research institute in paper 04, and tightened several framework mappings. The guide was updated for Colorado’s new AI law (SB 26-189).

These papers are general education, not legal advice. Rules change and specifics vary.