OpenSemantics Resource Library

SADAR Addresses Governance Gaps Identified in the FINOS AI Governance Framework for Discovery and Authorization

Maps SADAR's specification surface against the FINOS AI Governance Framework v2.0 agentic risk catalogue and the fourteen CVEs disclosed in the OX Security MCP supply chain research. Of the six FINOS agentic risks, SADAR structurally closes two, partially addresses three, and leaves one out of scope. Of the five CVE families analyzed, SADAR closes STDIO command injection and marketplace supply chain compromise architecturally. The analysis is honest about boundaries: SADAR supplies the discovery, attribution, and identity primitives that authorization engines need, but does not replace runtime observability, policy decisions, or network-layer enforcement.

What is SADAR?

Agents are only as effective as the capabilities they can discover, understand, and trust. Current approaches rely on predefined tool lists or tightly coupled integrations that do not scale to open, dynamic ecosystems. SADAR introduces a discovery layer focused not on execution but on semantics and trust — letting agents understand what a capability does, evaluate whether it meets their intent, and assess whether it can be trusted. The shift is from interfaces to intent: instead of asking "which API should I call," agents ask "which capability satisfies this objective." Identity, non-functional requirements, and compliance metadata become first-class selection criteria, not afterthoughts.

Trustworthy AI is Not a Feature - It is a Prerequisite

Trust in regulated industries — aviation, pharmaceuticals, healthcare, finance — is not earned through capability claims but through five demonstrable controls: responsible decisions, protection from misuse, controlled execution, monitoring against expected outcomes, and remediation of root causes. Agentic AI breaks the deterministic assumption these controls were built on. Agents do not execute predefined logic; they infer actions from context, making the same task reproducible in outcome but not in path. The control gap is not a gap in policy intent — it is a gap in infrastructure. SADAR provides the semantic foundation that makes those traditional five controls achievable in agentic environments.

The Hidden Infrastructure Crisis of Agentic AI: Why We Need a Semantic Registry

The real promise of Agentic AI requires that agents be able to broadly discover other agents, tools and resources such as files and databases. Today's implementations constrain this discovery by providing defined lists of things that can be discovered and relying on the agent to 1 - Identify the need to use them 2 - Understand the order in which they must be used 3 - Successfully match data - both in business meaning and syntax - as inputs and outputs This works for pre-defined, low-risk flows but falls short of the vision of agents independently performing tasks. For this, we need to define not agents and tools but capabilities and data in business in a Semantic Registry.

The Missing Agentic Infrastructure

Agentic AI promises autonomous systems that can discover, reason, and act—but the infrastructure required to make that safe and reliable doesn’t exist yet. Today’s agents can execute tasks, but they cannot correctly discover, interpret, or govern interactions with tools and other agents. Until we solve this infrastructure gap, agentic AI will remain confined to controlled demos rather than real-world enterprise use.

The Compliance Wall: When Black-Box Agents Meet Regulated Realities

Agentic workflows currently operate without verifiable attribution or transparent execution records, making them inherently non-compliant with frameworks like SOC 2, HIPAA, and FedRAMP. This article explores how autonomous systems violate the deterministic assumptions of traditional audits, and why post-execution logging fails when you cannot reconstruct the chain of decisions that led to an agent's action.

The Architecture Choice You're Already Making

Many agentic implementations have avoided an agentic AI governance crisis by deploying to low-risk and/or constraining agents to tightly scoped and deterministic use cases that fall short of the vision of fully autonomous agents. These constraints are not a substitute forgovernance – they are a barrier to value.

Why do we need another registry? Discovery, Not Lookup

API, AI, DNS, and other gateways/firewalls/registries support looking up something already known to the requester. Using a well-known name, they return technical information about the execution mechanics. That's important but agents start with "What should I use?". Given a task or goal, the agent must determine what other agents, tools, and/or resources are available and support achieving the task. This is, fundamentally, a different problem than today's registries address.

Why MCP and A2A Need a Semantic Layer

Today's frameworks focus on execution - the transport and data exchange. They require prior knowledge of agents, tools and resources organized in a curated list for the requesting agent. Even then, studies show agents aren't reliable in deciding if, how, or when to use agents and tools.

SADAR Quick Reference

SADAR is an open Community Specification aimed at improving the observability, attribution, and trust of Agentic AI systems.

Agentic AI Infrastructure Gap

Agentic AI promises autonomous systems that can discover, reason, and act—but the infrastructure required to make that safe and reliable doesn’t exist yet. Today’s agents can execute tasks, but they cannot correctly discover, interpret, or govern interactions with tools and other agents. Until we solve this infrastructure gap, agentic AI will remain confined to controlled demos rather than real-world enterprise use.

MCP: Fifteen CVEs, Two Protocols, One Missing AI Architectural Layer

Fifteen Agentic CVEs aren’t isolated flaws—they’re symptoms of a deeper architectural gap. Today’s agentic systems can decide how to act, but not whether they should. Until that missing layer is addressed, true enterprise adoption of Agentic AI will remain out of reach.

SADAR Executive Briefing

Even the perfect agents will misbehave without the right architectural controls. That's missing from today's agentic AI platforms and standards. Without these controls, agents cannot scale to their potential value and organizations are at risk.

The Agentic Tower of Babel

The original Tower of Babel collapsed because no one could understand anyone else. We are about to build the agentic version — thousands of agents, each fluent in its own private vocabulary, asked to discover, compose, and transact with each other at runtime. The fix is older than the problem, and it's already sitting in plain sight.

Why an Identity Card Isn't Enough: The Hidden Governance Crisis in Agentic AI

Assigning an AI agent an identity only verifies who the agent is, failing to account for its operational intent, business context, or authorizing human. To safely deploy autonomous systems, enterprises must implement a semantic infrastructure layer that enforces the originating scope of authority, tracks specific transaction instances, and evaluates the business state of the data being processed.

The 10% Reality: Why Agentic AI is Stalling in the Enterprise

Despite general AI adoption reaching over 80%, scaled deployment of autonomous AI agents remains stuck in the single digits. This article explores why the barrier isn't model intelligence, but rather a lack of enterprise trust. We discuss how probabilistic models break traditional deterministic business controls, and why "the agent did it" is not an acceptable answer for high-stakes workflows.

SADAR and SPIFFE: Two Specs That Complement One Another

A look at how SPIFFE's workload identity exists under SADAR's Agentic Capability IDs seamlessly. SPIFFE tells you where workloads can run while SADAR tells you what Agentic capabilities do, under what originating authority, and as part of what process providing deeper control and attribution.

The ROI of Accountable AI: Breaking the Control Barrier to Unlock True Autonomy

The real value of Agentic AI is locked behind the compliance gap barrier. Today's standards and infrastructure don't support the robust agent discovery, attribution, and grounding necessary for business-critical processes.