Writing
Long-form, evidence-led writing on AI runtime security: how agents are attacked, how containment actually behaves, and what buyers should demand before they trust a control. Every claim here describes what the product does today, and outside research is cited as reported rather than restated as our own finding.
In the spec-driven era a developer's value shifts from coding speed to precision of intent and rigour of verification. Includes reported Faros AI data on review time in high-AI-adoption teams.
Read the articleWhat escape actually means for AI agents, why swarms spread an injected instruction, why layered encryption is not containment, and how the triple-layered sandbox responds.
Read the articleUX, DX and now AX. Being cited in model answers rather than ranked, documentation an agent can operate, and why ease of use is a security question.
Read the articleThe anatomy of direct and indirect prompt injection, why model-side alignment is not a control, and what a deterministic policy layer catches before a model is reached.
Read the articleEach OWASP LLM risk set against the control that actually mitigates it at runtime, and an honest note on the ones a proxy cannot solve.
Read the articleWhat an auditor asks for, why a written policy is not a control, and how tamper-evident decision records turn governance claims into evidence.
Read the articleThe questions to put to any AI security vendor, including the ones that separate a demonstration from an enforced control.
Read the articleAccountability lines for autonomous systems, and the records you need before an incident, not after it.
Read the articleWhy subsidised model economics push control away from the buyer, and why the enforcement point is the part an enterprise should own.
Read the articleThe recorded demo walkthrough shows real traffic being refused and a guardbot dropping a containment layer by itself, and the Shield console lets you build a policy and test it in the same page.