Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

Anthropic's Inference Hooks update lets security teams validate Claude's tool calls before they run. Zenity supports it across Claude's surfaces, so teams can approve write-capable connectors, Model Context Protocol (MCP) servers, and wider Claude Code rollouts with policy on each action.

Public discussion of AI-enabled cybercrime tends to lean heavily on vendor marketing and speculation about future capabilities. We wanted to measure what criminal practitioners are actually discussing, what tools they prefer, and where their attention is shifting.

AI agents are multiplying faster than most organizations can track, let alone govern. IDC predicts there will be 1.3 billion agents by 2028. Every team that adopts one is also, often without meaning to, creating a new identity. And it’s one that can access systems, move data, and act on someone's behalf. The rush right now is to get ahead of that before it spirals, and the first step is to answer a basic question: whose job is this?

Interaction monitoring covers four of the ten risks in the 2025 Top 10 for LLMs and generative AI applications well, three partially and three not at all. The split is not arbitrary, and the line it follows tells a buyer more than the coverage count does. ‍ Monitoring observes systems being used. Several entries on the list describe defects in how a system was built, which no amount of observation at the point of use will surface.

Search for agent risk quantification tools and most results are risk assessment tools. Questionnaires, maturity scores, compliance trackers and templates, all producing an ordinal rating. A rating ranks one agent against another. It does not tell anybody what to spend, what to insure or what to stop. ‍ The separating question is whether the output arrives in currency.

A coding agent on a laptop runs as the developer. It reads the same keychain, carries the same tokens and appears in every downstream log under the same name. The identity record is accurate and it cannot answer the question governance needs answered. ‍ Every control keyed to identity therefore operates on a merged subject.

A security leader finishes a pitch to the board about an agentic SOC that pairs AI agents with analysts to address the coverage gap. A hand goes up. “So, the AI decides who’s a threat?” It’s a fair question. It’s also the wrong one, built on a myth that’s common enough to stall adoption on its own.

In the first post in this series, Wade shared how our team laid the foundation for using AI in detection engineering. The short version is that an LLM doesn't know your stack. Without documentation and workflow context, it can produce plausible results that are difficult to trust, regardless of which model you use.