The security and governance disciplines around AI are not the same discipline, and the productivity costs of confusing them are real. Security is the classical practice — keeping attackers out, limiting their damage when they get in, monitoring for the breach you didn’t prevent. Governance is something different — keeping the organization’s own use of AI from causing damage to itself: deploying agents that act on production systems without oversight, surfacing PII through a chat interface that wasn’t supposed to see it, propagating a small mistake into a thousand customer emails before anyone notices.
This post is the field guide to both, organized as the natural sequel to How Hackers Weaponize AI. That post covered the attacker’s side; this one covers the defender’s side and the governance disciplines that defenders are increasingly responsible for. Together they form a two-piece series on AI security and risk.
What it covers
About thirteen minutes of reading.
§ 1 — The 2026 cybersecurity landscape. What’s changed in the last twelve months. The AI-enabled attack surface. The new defender-side AI tooling. The handful of trends actually worth tracking versus the dozen marketed as trends but not.
§ 2 — The security vs. governance split. The clear definitions. Why both matter. The org-design implications (whether they should report to the same person — usually no).
§ 3 — Architecting secure AI agents. The handful of patterns. Capability scoping. Tool-call validation. The principle of least privilege applied to LLM tool surfaces. The reversibility audit (what actions can the agent take that cannot be undone?).
§ 4 — Zero-trust for autonomous agents. Extending the zero-trust model to agents. Authenticate every tool call. Verify every action. Default-deny on operations that hit production data. The implementation patterns.
§ 5 — Governance disciplines. The non-security half. Audit logs for agent actions. Human-in-the-loop placement. Approval workflows for high-impact actions. The kill switch and the kill-switch-test discipline.
§ 6 — The five IBM Technology talks that inform this. Sources: Cybersecurity Trends 2026 · Security & AI Governance · Architect Secure AI Agents · Securing & Governing Autonomous Agents · Securing AI Agents with Zero Trust. Synthesized into one coherent picture.
§ 7 — A practical baseline. The minimum disciplines a 2026 organization should adopt. Twelve items, ordered by impact-per-effort.
§ 8 — Further reading. The NIST AI Risk Management Framework. The OWASP Top 10 for LLMs. The MITRE ATLAS taxonomy. The handful of academic papers that have actually changed practice.
Read it
The field guide lives at its own URL with a warm-paper layout — deep teal as the dominant accent, amber for warnings. Eight sections plus a practical baseline checklist.
← Back to Autonomy