# dibOS — Competitive Analysis: Source Note **Prepared:** October 2026 **Companion to:** `index.html` (the analysis dashboard) **Method:** 10 parallel browser research agents, one narrow question each **Status of dibOS evidence:** none retrievable in public sources --- ## 1. Evidence boundary No public footprint for dibOS was found. Searches for `dibOS`, `dib os`, `dibos.ai`, `dibos.io`, `dibos.dev`, "dibOS governance", "dibOS authority engine", "durable institutional boundaries", and `dibos` on GitHub returned no product, documentation, pricing page, funding announcement, launch post or repository belonging to this product. The domain `dibos.ai` does not resolve (DNS failure). **Consequence:** the analysis evaluates the *thesis* — "durable institutional boundaries operating system / authority engine for humans and AI agents" — against the market as it exists. Every dibOS score in the scorecard is a thesis score, not an observed capability. Replace with internal material before using in a fundraise or board paper. --- ## 2. Findings by research thread ### Thread 1 — dibOS footprint No findings. Domain checks and searches returned nothing attributable to the product. ### Thread 2 — Microsoft agent identity and governance - Four new directory object types: agent identity blueprint, agent identity blueprint principal, agent identity, and agent user. A blueprint can create one or more agent identities, optionally with an agent user for each. - "Agent identities are accounts within Microsoft Entra ID that provide unique identification and authentication capabilities for AI agents." - Agent identities inherit OAuth 2 delegated permission scopes from their parent blueprint. - Human accountability: "Sponsors of agent identities are human users accountable for making decisions about its lifecycle and access," alongside owners and managers. - Automatic provisioning: Microsoft Foundry provisions a default blueprint and default agent identity per project; publishing an agent creates a dedicated blueprint and identity; Copilot Studio can auto-create Entra agent identities (preview); Teams agents get blueprints in the Developer Portal for Teams. - Access packages (Entitlement Management) grant agent identities security group memberships, OAuth API permissions (delegated and application, for Graph and applications) and Entra roles. Assignment policies support 1, 2 or 3 approval stages plus an expiry date. - Three request pathways: the agent itself programmatically (`accessPackageAssignmentRequest`), the agent's sponsor on its behalf, or an admin direct-assign. - Stated limitation: agent identities and service principals "can't be added through access packages to application roles, SAP roles, or SharePoint Online site roles," so existing access packages containing those cannot be reused. - Conditional Access for agents evaluates agent context and risk, including agent identity risk from Entra ID Protection, and can be applied at blueprint level so all identities inherit it. - Lifecycle: time-bound access with expiry; sponsor receives pending-expiration notifications and can request extension, which can trigger a new approval cycle; otherwise the assignment "automatically expires on its end date, and the agent identity loses access." Sponsors and owners can enable and disable agents from the My Account and My Access portals. - Agent 365 exists: "a Microsoft 365 service that provides a centralized control plane for AI agents. It enables organizations to discover, manage, govern, and secure agents across their environments," covering Microsoft-built and third-party agents. - "Microsoft Entra Agent ID provides the identity platform capabilities for Microsoft Agent 365." - Agent 365 capability names: Registry, Agents map, Agent analytics, Role-specific oversight, Agent onboarding (IT-controlled workflows with security policy templates), Integration management (least-privilege control over which users, data, tools or MCP servers agents can use), Lifecycle management (rules-based, e.g. expiring inactive agents). Admin surfaces: Microsoft 365 admin center, Defender, Entra, Intune, Purview. - **Pricing:** Agent 365 **$15.00 per user/month, paid yearly (annual commitment)**; Microsoft 365 E7 **$99**; E7 without Teams **$90**. "Agent 365 is licensed per user. An Agent 365 license is recommended for all users who interact with, own, manage, or sponsor Agent 365–managed agents. Agent 365 is also included with Microsoft 365 E7." - Governance licensing: Entra ID Governance for agent identities requires Microsoft 365 E7 (which includes Agent 365 and Microsoft Entra Suite), or Agent 365 paired with at least Entra P1 or Microsoft 365 E3. - **Unfinished:** the "Microsoft Agent 365 Licensing FAQ" page could not be read (Learn paths returned not-found), so any per-agent metering or agent-count caps are not captured. No hard numeric limits (max agent identities per blueprint or tenant, token/call quotas) were found on any page reached. The "Management of agents" section and detailed audit-log specifics were not fully read. ### Thread 3 — Okta / Auth0 delegated authority - Okta Cross App Access (XAA): the requesting app or agent obtains an ID-JAG (Identity Assertion JWT Authorization Grant, an OAuth 2.0 extension) after the user signs in; the external authorization server validates it and issues a short-lived, scoped access token; the client uses it to access the protected resource on the user's behalf without repeated consent prompts. XAA supports authorization chaining across domains. - Okta Agent SSO, Agent as Principal, On-Behalf-Of and Auth for MCP extend the model to agent flows; Auth0 for AI Agents positions these as removing custom identity work that stalls agent deployments. - **Unfinished:** detailed pricing for the agent-specific SKUs was not captured. ### Thread 4 — External authorization / policy decision points - Cerbos: "Conditions are CEL expressions evaluated at request time against principal, resource, and request attributes. A rule with a condition applies only when the expression evaluates to true." Spending limits and approval thresholds are therefore expressible as conditions, but quorum, escalation and separation of duties must be modelled by the customer. - OpenFGA / SpiceDB: relationship-based access control; organisational hierarchy and desk ownership are expressible as relationship graphs. - **Unfinished:** per-vendor pricing tiers and self-hosted vs SaaS terms were only partly captured. ### Thread 5 — AI governance / compliance platforms - Across six platforms reviewed, **none is primarily a runtime-authority enforcement layer for agents**. Four — Credo AI, IBM watsonx.governance, FairNow, Trustible — are inventory, policy and evidence platforms with EU AI Act documentation. Two — Holistic AI and OneTrust — claim actual runtime intervention on AI systems and agents. - Credo AI modules: AI Registry & Discovery, Unified AI Governance Platform, Governance Knowledge Graph, Governance AI Agents, modular by design. - **Unfinished:** pricing for all six was not captured; comparison literature was used for scope. ### Thread 6 — ServiceNow AI Control Tower and suite incumbents - AI Control Tower is sold as a **package**, not a per-seat list price (page labels it "PACKAGE — Super Intelligence Control Tower"). - Built on the CMDB and CSDM 5.0; discovers and inventories AI assets (internally built, third-party, agent-driven); layers risk and compliance management, lifecycle management, third-party AI risk and tamper-evident audit logging; maps to NIST AI RMF and EU AI Act Article 12; wired into IRM and ITSM. - Framework coverage documented as partial across EU AI Act, NIST AI RMF, HIPAA and GDPR Art. 22. - **Unfinished:** Salesforce Agentforce, Workday Illuminate and SAP agent governance were not reached. ### Thread 7 — Durable execution - DBOS: durable workflow orchestration from MIT-Stanford research; makes applications and AI agents durable, fault-tolerant and scalable via annotations on ordinary code. Open source. - Temporal, Inngest, Restate: comparable durable-execution positions; 2026 comparison literature is explicitly framed around "durable execution for AI agents." - **Unfinished:** per-vendor pricing and any approval-gate primitives were not captured in detail. ### Thread 8 — Runtime guardrails and observability - Arize AX "Guardrails" claims genuine runtime enforcement: "Guardrails correct undesirable outputs at run-time, ensuring real-time safety and compliance. Failed messages trigger corrective actions such as default responses, retries, or blocking outputs entirely." Guards intercept the LLM call and can act on distance thresholds. - Others in the set (Galileo, Prompt Security, WitnessAI, Lasso, Noma, Protect AI, LangSmith) split between in-line blocking and post-hoc scoring; the block-vs-observe distinction is the key discriminator. - **Unfinished:** per-vendor enforcement verification and pricing were partly captured. ### Thread 9 — Standards and regulatory drivers - NIST NCCoE concept paper, 5 February 2026: "Accelerating the Adoption of Software and AI Agent Technologies" — explicitly asks about identification, authorization, auditing and non-repudiation of AI agents, plus supporting controls. - OASIS "Inside AI Governance White Paper 2026" documents the regulatory landscape, including a Brookings Institution January 2026 analysis of 385 state AI bills finding legislative activity tracks ideology and fiscal conditions. - Agentic AI market forecasts for 2026 range from roughly $7B to $201.9B across four firms, with one estimate at $9.14B growing at 42% CAGR — the spread itself indicates an immature category. - **Unfinished:** Singapore IMDA Model AI Governance Framework for Agentic AI and the BIS/FSI report on AI in financial services were not read. ### Thread 10 — The authority school of thought and its peers - Dock Labs, "How we designed verifiable chains of authority for AI agents" (Sep 18, 2026): agents can "prove who authorized them, what they can do." - Helixar Labs, HDP — Human Delegation Provenance Protocol: an open standard for recording and verifying human delegation to agents. - ElixirData, "Delegation Chains: Who Authorized the Agent to Do That?" (Sep 28, 2026). - Agent Authority Chain: "Put enforceable limits on every AI agent. Create a free organization account, verify your domain, install an official integration…" - AgentBound, "Verifiable Behavioral Governance for Autonomous AI," arXiv 2606.30970v1 (Jun 29, 2026). - COMPEL Framework, "Agentic AI Governance Architecture: Delegation, Authority, and Accountability" (Apr 6, 2026). - CERTEN: "The Control Layer of Digital Finance," "Permissioned control for every asset on every chain." Badges: Private Testnet Live, 7 EVM testnets live, 6 more code-complete, 50+ API endpoints. - Accumulate whitepaper v1.0 (12 Apr 2022), DOI 10.6084/m9.figshare.19586515, Inveniam DeFi Devs. - "Provable Institutional Control for Tokenized Finance," Jay Smith, SSRN abstract_id=6864258. - Beige Media, "Governing AI Agents Through Institutional Offices" (Jul 31, 2026) — the institutional-office framing that dibOS's language most closely echoes. - **Unfinished:** a full read of the Beige Media piece and the CERTEN whitepaper's control primitives. --- ## 3. Scorecard method Weights: institutional semantics 20%, durable record 15%, enforce at action 15%, human+agent symmetry 10%, cross-org 10%, audit grade 10%, deployment sovereignty 8%, time-to-value 7%, distribution 5%. Scores 1–5 per dimension, weighted sum normalised to a 5-point scale. Scores are analytical judgements from the evidence above, not vendor benchmarks. dibOS's row is a design target for the thesis, not an observed result. --- ## 4. What is verified vs. asserted **Verified (read on a cited page):** Microsoft object types, access-package approval stages, sponsor roles, Agent 365 pricing ($15/user/month, E7 $99), the SAP/SharePoint access-package limitation, Okta XAA mechanics, Cerbos CEL conditions, ServiceNow package-only pricing and partial framework coverage, Arize runtime guardrail claims, the NIST NCCoE concept paper's existence and scope, the existence and titles of the Dock, Helixar, Agent Authority Chain, CERTEN, Accumulate, COMPEL and AgentBound materials. **Asserted, not verified:** competitor pricing outside Microsoft; enforcement claims by Holistic AI, OneTrust and the guardrail vendors; the completeness of any vendor's runtime coverage; the absence of a dibOS product (absence of evidence, not evidence of absence). --- ## 5. Revisions - v1.0 — October 2026. Initial analysis from 10 parallel research agents; 7 returned partial results at the fleet spend ceiling. Unfinished threads are listed per thread above.