Pervaziv AI Adds Decision Intelligence to Cortex with Two New Specialized Models
Routing Decisions and Security Decisions bring sub second routing, faster security triage and a new decision layer to the expanding Cortex AI Model Ensemble.
San Francisco, CA October 19th, 2026
Pervaziv AI today announced Cortex Routing Decisions and Cortex Security Decisions, the 11th and 12th specialized models in the Cortex AI Model Ensemble, introducing a dedicated decision intelligence layer for fast, structured routing and security judgments across the Cortex Enterprise AI Control Layer.
The release advances a simple operating principle: not every enterprise AI question needs a large generative model. Some moments require deep reasoning or evidence. Others require a precise decision about what should happen next.
Cortex Routing Decisions is designed for the second category. It helps determine whether a request should move to a supported action, coding workflow, connected information path, security capability, clarification step or broader Cortex workflow. Cortex Security Decisions applies the same specialized approach to recurring security judgments, including whether code appears vulnerable, the likely security category and severity, whether exploitation conditions appear present, and whether a proposed patch satisfies an initial security validity check.
Together, the models allow Cortex to use fast decision intelligence for common bounded judgments while preserving Cortex Router, Cortex Analysis, Cortex Verify and other specialists for work that requires richer reasoning, evidence, remediation or independent verification.
The result is a more selective AI system designed to spend deeper intelligence where deeper intelligence adds value.
Faster decisions. Deeper intelligence when it matters.
A New Decision Layer for Enterprise AI
Enterprise AI systems encounter many operational questions before generation or execution begins. Is a request asking for a direct action or a broader engineering workflow? Is the action sufficiently specified? Does the work require security review? Does code appear vulnerable? Is a proposed patch clearly invalid, apparently safe, or uncertain enough to require deeper verification?
Those questions can determine the path of an entire workflow, but they do not always require an open ended generative response.
Cortex Decisions adds a specialized layer for these moments. The models return bounded judgments that the wider Cortex control layer can use to select, validate, escalate or continue the appropriate path. They are intended to reduce unnecessary model calls without weakening the standard applied to consequential work.
This expands the Cortex AI Model Ensemble into three complementary forms of intelligence.
Local models support eligible private and responsive work close to the user and workspace.
Agent and backend models provide deeper reasoning, planning, coding, investigation, analysis, execution and verification.
Decision models provide fast structured judgments for routing, risk classification, security triage and early workflow gates.
The categories work together, with Cortex coordinating the level of intelligence appropriate to each stage.
Cortex Routing Decisions: Moving from Intent to Action Faster
Routing sits near the beginning of an AI interaction. Delay at that stage is experienced before analysis, planning or execution can even begin.
Cortex Routing Decisions is designed to reduce that delay for supported requests. It evaluates the operational form of an incoming request and provides a structured judgment about how Cortex should handle it. A precise workspace action can move toward the appropriate supported capability. A broader engineering request can enter a coding or planning workflow. A request that depends on connected information can follow that path. A security sensitive request can be directed toward the appropriate security capability. An underspecified request can be identified for clarification or broader investigation rather than being forced into an action.
The model complements the broader Cortex routing architecture. It does not eliminate the established Cortex Router path. For eligible requests, it can replace an unnecessary heavyweight generative routing pass. When a result is unsupported, incomplete, ambiguous or outside the fast path, Cortex can continue through its established routing workflow.
That distinction is central to the release. Cortex is not replacing depth with speed. It is replacing unnecessary depth with a specialized decision when the workflow does not require a larger reasoning pass.
In a controlled same host comparison across a broad held out routing suite, Cortex Routing Decisions completed the warm median routing stage in less than 100 milliseconds. That represented an approximately 20 times improvement over the existing Cortex Router comparison path and about 95 percent lower median latency.
Measured request throughput reached about 24 times the baseline. Resident accelerator memory was more than 60 percent lower than the Router comparison path. Route accuracy and combined route and action correctness were approximately 96 percent across the measured suite.
Faster routing reduces delay before useful work begins, while lower memory demand leaves more accelerator capacity for specialist workloads.
Cortex Security Decisions: A Faster Front Door for Security
Security workflows contain repeated classification decisions before and between deeper investigations. Teams may need to distinguish a likely vulnerability from a benign pattern, identify a probable security family, prioritize by severity, assess whether exploitation conditions appear present, or determine whether a proposed patch should be rejected, escalated or sent forward for verification.
Cortex Security Decisions is designed for those bounded judgments.
Cortex Security Decisions complements the deeper capabilities of Cortex Analysis and Cortex Verify. When teams need detailed code findings, remediation context, supporting evidence or comprehensive review, Cortex can continue to the appropriate specialist. The new model provides a faster initial classification for well defined security questions, while cases requiring deeper assessment or validation follow the established analysis and verification workflow.
In the controlled same host comparison, Cortex Security Decisions was evaluated against shared classification responsibilities from separate held out Cortex Analysis and Cortex Verify suites.
Against Cortex Analysis, the decision path delivered about 24 times faster median performance, corresponding to about 96 percent lower median latency. Measured throughput was about 22 times higher. Resident accelerator memory was more than 60 percent lower, and estimated energy use was about 97 percent lower for the measured decision workload.
Against Cortex Verify, Security Decisions delivered about 11 times faster median performance, with more than 90 percent lower median latency. Measured throughput was about 13 times higher. Resident accelerator memory was again more than 60 percent lower, with an estimated 94 percent reduction in energy for the measured workload.
The quality comparison is equally important. On shared labels in the measured suites, Security Decisions improved vulnerability and severity classification over Cortex Analysis while maintaining strong security family classification. It also exceeded Cortex Verify on the shared Boolean patch validity measure.
These comparisons are intentionally limited to responsibilities the models share. They do not measure or replace the richer evidence, explanation and verification records produced by Cortex Analysis and Cortex Verify.
A Selective Fast Path, Not a Security Shortcut
Performance alone does not determine whether a decision should bypass deeper analysis.
Cortex applies the new models selectively. In the held out comparison suites, carefully selected fast path cases avoided downstream Analysis and Verify calls with no observed errors in those selected paths. Uncertain or apparently valid patches continued to Cortex Verify, while severity decisions lacking sufficient impact context escalated to Cortex Analysis.
The objective is to resolve eligible bounded decisions efficiently while preserving deeper review for uncertainty, evidence requirements and consequential changes.
When remaining requests are charged for both the initial decision and specialist escalation, the projected hybrid path still produced material latency savings for Analysis oriented and Verify oriented traffic in the measured evaluation.
Benchmark results remain workload, configuration and hardware dependent. Energy figures are directional operating estimates derived from the measured processing and resource profile rather than universal power guarantees.
"Enterprise AI should not spend heavyweight intelligence on every small decision," said Anoop Jaishankar, Founder and CEO of Pervaziv AI. "The real opportunity is to match the intelligence to the responsibility. Cortex Decisions gives the Ensemble a faster control layer for routing and security while keeping deeper reasoning, evidence and human oversight exactly where they belong."
Jaishankar continued, "Speed is only useful when the system knows when not to take the shortcut. A fast routing judgment should not invent intent, and a fast security classification should not become an unsupported approval. We are building Cortex so routine decisions can move quickly, while ambiguity, risk and consequential changes still reach the specialists and controls designed for them."
Performance That Compounds Across the Workflow
The value of a faster decision layer compounds because routing and security classification often occur before more resource intensive stages.
A routing decision completed in under 100 milliseconds can begin the next stage almost immediately. Higher throughput increases control capacity, while lower resident memory demand can preserve capacity for Cortex Planner, Cortex Code, Cortex Analysis, Cortex Verify and other specialists.
For developers, that can mean less waiting before Cortex understands what kind of work is being requested.
For security teams, it can mean faster triage of common classifications while evidence rich and uncertain cases continue to deeper review.
For engineering leaders, it creates a clearer progression from intent to decision, planning, execution and validation.
For platform teams, it offers a path to improve utilization by reserving larger model capacity for tasks that genuinely require generation, investigation and verification.
This reflects the broader Cortex direction: capability through clearer responsibilities across specialized intelligence.
From Routing and Planning to Decisions
Pervaziv AI introduced Cortex Router as a coordinating entry point for specialized AI and later expanded routing across models, search and engineering skills. Cortex Planner added structured execution planning for complex engineering work.
Cortex Routing Decisions and Cortex Security Decisions add another layer between intent and execution: fast judgment.
A request can enter Cortex, receive a routing decision, pass through security decisions when risk is relevant, move into planning when work is complex, and reach deeper specialists for implementation, analysis or verification. Policy, permissions, validation and human review remain part of the surrounding control layer.
The progression can be short for simple requests and deeper for consequential ones.
Replacing Model Calls, Not Model Depth
For eligible requests, the new decision models can replace selected model calls. They do not remove the specialist models from the Cortex AI Model Ensemble.
Cortex Routing Decisions can replace a slower generative routing pass when the route is supported and validated.
Cortex Security Decisions can replace an Analysis or Verify call when the workflow requires only a bounded classification and the applicable controls permit the result to be used.
Ambiguous requests, detailed security findings, remediation guidance, complex execution, uncertain classifications, consequential patch approval and evidence requirements continue to the appropriate specialist path.
That separation gives Cortex a more intentional way to allocate intelligence across the workflow.
The 11th and 12th Models in an Expanding Ensemble
Cortex Routing Decisions and Cortex Security Decisions extend the Cortex AI Model Ensemble to 12 specialized models and establish decision intelligence as a distinct category within the platform.
The direction remains consistent with Pervaziv AI's broader vision for Cortex as an Enterprise AI Control Layer: connect user intent with the appropriate intelligence, context, permissions, security, execution and validation while keeping the experience coherent across supported work surfaces.
Local intelligence for private and responsive work. Agent and backend intelligence for deep reasoning and execution. Decision intelligence for fast governed control.
The goal is not simply more models.
It is to make every model call more intentional.
Availability
Cortex Routing Decisions and Cortex Security Decisions are being integrated into the Cortex AI Model Ensemble for eligible routing and security workflows. Availability and workflow behavior may vary by Cortex experience, deployment configuration and applicable organizational controls.
About Pervaziv AI
Pervaziv AI develops Enterprise AI and cybersecurity technology for modern software teams. Cortex brings together specialized AI models, coding, security, privacy, connected enterprise context, browser and developer workflows, cloud execution, validation and governance within an Enterprise AI Control Layer designed to help organizations build, secure and deliver software with greater speed, visibility and control.