Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

When an AI system misfires inside a large company, the question that follows is who was responsible for it. The answer usually lands on the CIO, and frequently the CIO had no part in choosing, approving or deploying the thing that failed. ‍ This gets described as a governance problem, which undersells it. Governance implies a process that exists and performs badly.

Somebody in the organization will eventually propose running documents through an AI text detector. The instinct is reasonable and the method is the one approach the published evidence rules out, for reasons a security or risk function should understand before the policy gets written. ‍ The organization that built the most widely used language model tried this and withdrew its own tool after six months. Its reason was accuracy, and the figures it published before withdrawing explain why. ‍

Every major identity vendor shipped an agent offering during 2026, and the datasheets converge on the same three claims. Agents get first-class identities, permissions are scoped, and access is governed. ‍ The question that separates them is narrower. When the engineer who built an agent leaves the company, what happens to the agent? The answers differ enormously, and they differ for reasons that trace back to what each vendor was originally built to do. ‍

A team that can write a Monte Carlo simulation over a loss distribution can build a cyber risk quantification model in a few weeks. The mathematics is undergraduate statistics and the code is not the obstacle. ‍ Three other things are, and each one outlasts the build. Loss data good enough to calibrate against, the work of keeping the model current once the person who wrote it moves on, and documentation that holds up when somebody who did not build it starts asking questions. ‍

LiteLLM, a widely used open source gateway for routing requests across language model providers, was compromised twice in 2026. Malicious releases reached its public package registry in March and again in October, through two separate intrusions, each carrying a credential stealer into every environment that installed them. ‍ Neither attack touched a model. Neither one manipulated weights or tampered with training data.

Quantification integrated with a governance platform and quantification sitting beside one behave nothing alike after the first quarter. The integrated figure moves when control status moves. The standalone figure describes the date somebody last updated it. ‍ Two directions of data traffic make the difference, and most implementations build only the first one. ‍

Microsoft states the position plainly. Copilot only surfaces organizational data to which individual users have at least view permissions, so it grants nobody access to anything they could not already open. What changes is that material somebody had permission to reach and never would have found now arrives in a summary. ‍ The governance work is therefore permissions work that predates the deployment.

Measure carries four categories and twenty-two subcategories, and they divide cleanly by what kind of evidence satisfies them. Some ask whether something was observed. Others ask whether something was tested. Monitoring answers the first group and contributes almost nothing to the second.

Buying a cyber risk quantification platform means asking a vendor to defend numbers that will end up in front of a board, an auditor, or a regulator. Most evaluation calls never press that hard. The questions below are the ones that separate platforms from each other, and they are written so that any vendor can answer them, including vendors who would beat Kovrr on several of the answers.

Conventional vendor risk tooling asks whether a supplier will suffer a breach. For an AI provider that question is necessary and incomplete, because the larger exposure is what the provider does with your material while operating exactly as intended. ‍ Retention periods, training use, inference location and the model serving an endpoint are data governance questions rather than security ones. Most platforms in this market have no field for them, which is the first thing to check. ‍