Two announcements dropped this week that, read together, describe a single structural problem in enterprise security: the gap between what AI-powered tooling can automate and what the underlying environment has documented well enough to be automated against. One vendor is shipping agentic SOC workflows that execute detection, investigation, and onboarding tasks at machine speed. Another is shipping a tool whose entire value proposition is answering a question most large organizations cannot currently answer: where is your cryptography? The combination is instructive, and not in the way either press release intends.
The Architecture
Anvilogic’s Blueprints capability, now generally available, is an orchestrator-agent model for security operations. The design is federated by intent: agents reach across Splunk, Sentinel, Snowflake, S3, Azure Blob, and other existing stores without requiring data centralization. Workflows are parameterized by customer configuration, including MCP connectors for tooling outside Anvilogic’s native integrations, and can be scheduled to run continuously against recurring detection tasks. The announced production result — a Fortune 100 telecommunications company onboarding data feeds in approximately 15 minutes each and eliminating a backlog of over 100 feeds — is the kind of concrete operational claim worth taking seriously. That is real work being automated.
QSE’s QScope operates at a different layer. It is a CBOM compiler: a tool for building a Cryptographic Bill of Materials by scanning an organization’s systems to surface which algorithms, certificates, keys, and libraries are actually in use. The driver is post-quantum migration. NIST has standardized post-quantum algorithms, U.S. federal agencies face a 2035 hard migration deadline, and the threat model behind all of it — harvest now, decrypt later — means the exposure window is already open for data with long retention lifespans. You cannot migrate cryptography you have not inventoried.
Why the Gap Matters
Here is the engineering problem that sits between these two systems. Agentic SecOps platforms like Anvilogic operate on telemetry, alerts, and detection logic. They are designed to move faster than analysts can, which means they inherit the completeness assumptions of whatever environment they run against. If your cryptographic inventory is incomplete — if you do not know that a legacy internal API is still terminating TLS with a deprecated cipher suite, or that a third-party library embedded in a containerized service is using RSA-2048 keys scheduled for deprecation — then no amount of agentic automation fixes that blind spot. Agents cannot detect what the underlying data does not surface.
This is not a criticism of the agentic model. It is a constraint that the agentic model makes more visible. When a human analyst investigates an alert manually, they can notice adjacent anomalies — an unexpected certificate, an unusual handshake, a configuration that does not match the baseline. When an agent executes a defined workflow against defined telemetry, it executes exactly that workflow. The scope of its awareness is the scope of what was instrumented.
The CBOM problem is a precondition problem. The reason cryptographic inventory is invisible in most organizations is that no single team owns it. Network teams own network devices. App teams own application dependencies. Cloud teams own cloud configurations. Certificate management often lives in a ticketing system that no one queries systematically. This is not negligence — it is an emergent property of how large organizations build systems over time. But it means the LLM-powered workflow orchestrator running your SOC is making decisions inside an environment whose security posture includes a category of risk that has never been formally documented.
What Actually Changes When Agents Run Security Operations
Automated agents operating at scale have a property that human analysts do not: they can act consistently on incomplete information without recognizing that the information is incomplete. A human analyst reviewing an investigation will occasionally ask whether they are missing context. An agent executing a Blueprint will close the workflow when the defined conditions are met. The organizational incentive structure that produced the undocumented cryptographic sprawl in the first place — diffuse ownership, no forcing function for inventory — does not change when you add an agentic orchestration layer on top.
What does change is the blast radius. Agents running continuously against a larger connected data surface, with MCP connectors reaching into tooling across the organization, have broader access than any individual analyst. If the threat model includes a compromised agent or a manipulated workflow — and it should, given that LLM-powered systems have documented prompt injection and context manipulation vulnerabilities — then the access scope matters enormously.
Author’s Position
The sequencing these two announcements imply is the right sequencing, even if neither vendor puts it that way. Cryptographic inventory is not a post-quantum project. It is a precondition for operating any security automation that touches encrypted traffic, certificate validation, or key management. You cannot write a detection rule, a response workflow, or an agentic Blueprint against cryptographic risk you have not documented. And you cannot understand the access surface of an agent-based SecOps platform without first understanding what that platform is being asked to reason about.
Practitioners deploying agentic SOC tooling should treat CBOM generation as a prerequisite, not a parallel workstream. Before you schedule a Blueprint to run continuously against your environment, establish what that environment’s cryptographic posture actually is. Not because quantum computers are breaking RSA this quarter — they are not — but because agentic systems that act on incomplete environmental models produce confident, automated responses to an incomplete picture of risk. The speed advantage of automation becomes a liability when the model it is acting on has structural gaps.
The precondition that makes this failure class possible is organizational: no team owns cryptographic inventory, so no team produces it. Adding AI-powered automation does not fix that ownership problem. It inherits it.
References
- You Cannot Protect What You Cannot See: A Cybersecurity Company Just Built a Map for the Encryption Hiding Inside Your Business
- Anvilogic Advances Agentic SecOps with the General Availability of Blueprints and a New Federated Search Experience, Powered by Anvilogic Compute
Perspectives
The part the vendor demo never shows you is the AI agent confidently patching a CVE in a cryptographic library that your inventory doesn’t know exists, because your inventory was built from the assets you documented, not the assets you have. Selling agentic SOC automation into an undocumented cryptographic environment is like deploying a smoke detector in a building whose blueprints are missing two floors — the detector works exactly as advertised, and the fire does too. A Cryptography Bill of Materials isn’t a nice-to-have that gets funded after the platform purchase; it’s the precondition for the platform doing anything other than automating your existing ignorance at machine speed. The official account says the AI closes gaps. What it actually does is find the gaps you already mapped and ignore the rest with impressive efficiency.
Agentic SOC platforms will do exactly what you built them to do, at machine speed, against the data you actually have — and if your cryptographic inventory is a fog of undocumented assumptions, you have automated yourself a faster way to miss the same things you were already missing. The steelman for “run the agents first, inventory later” is that you get *something* working while the cataloguing effort drags on for two years; the problem is that CBOM generation isn’t a refinement you bolt on afterward, it’s the input layer, and running detection workflows without it is operationally equivalent to a smoke alarm that only covers rooms you remember to list. The specific mechanism here is completeness: an agent’s detection chain is only as wide as the asset graph it queries, and undocumented cryptographic implementations are invisible to that graph by definition — not because the AI is bad at its job, but because no query returns rows that don’t exist. Build the inventory first, then let the agents run fast; in that order, you get compounding returns on both investments, and in the other order you get a very impressive dashboard of partial truth.
The failure here is organizational before it is technical: a SOC that cannot see its cryptographic surface has a collective knowledge problem, not an automation problem, and no amount of agentic velocity fixes a shared blind spot that the group never acknowledged it had. What gets automated is what gets counted, and what gets counted is what someone, at some institutional moment, decided was worth counting — that decision is a social fact embedded in the organization’s history, not a gap the next tooling purchase will reach. Groups systematically overestimate the completeness of their shared maps of the environment, a phenomenon well-documented in distributed cognition research, and agentic SOC platforms do not disrupt that dynamic — they accelerate inside it, running faster and faster across terrain the group collectively agreed, implicitly, to stop mapping. CBOM generation is not a precondition for the AI workflow; it is a precondition for the organization having an accurate model of what it is actually defending, and no workflow can substitute for that collective epistemic act.
Agentic SOC systems execute detection workflows faster and more consistently than human analysts, but speed applied to an incomplete inventory is not an advantage — it is a mechanism for producing wrong answers at scale. A cryptographic bill of materials is not a feature you add after deployment; it is the precondition for the system having anything accurate to reason about. An AI agent scanning an environment for cryptographic risk against an undocumented asset surface is not performing analysis — it is performing the appearance of analysis, which is measurably worse than acknowledged ignorance because it displaces the uncertainty rather than resolving it. The performance gap between human and machine decision-making in security operations closes to zero, and then inverts, precisely when the machine is operating on inputs that humans never bothered to enumerate.
