Andrii Bidochko

• Updated: August 22, 2026
• 6 min read

No One to Blame: A Framework of Constitutive AI Unaccountability

Direct Answer

The paper No One to Blame: A Framework of Constitutive AI Unaccountability introduces a diagnostic framework that identifies when autonomous, agentic AI systems are structurally incapable of being held accountable, regardless of policy or transparency fixes. It matters because the framework reveals hidden “accountability voids” that can render legal, ethical, and governance mechanisms ineffective, exposing organizations to systemic risk.

Background: Why This Problem Is Hard

Autonomous AI agents—ranging from large‑language‑model‑driven chatbots to self‑optimizing decision‑makers—are increasingly embedded in high‑stakes domains such as finance, healthcare, and critical infrastructure. Traditional accountability models assume three levers:

• Legal attribution: a clear human or corporate actor can be identified as responsible.
• Technical traceability: system logs and model provenance enable post‑mortem analysis.
• Normative oversight: standards, audits, and certifications guide acceptable behavior.

In practice, these levers break down when the sociotechnical configuration of an AI system obscures agency, diffuses responsibility, or embeds normative gaps that no amount of documentation can fill. Existing literature treats accountability gaps as “barriers” that can be patched with better standards or more transparency. That view overlooks configurations where the very architecture of the system makes accountability conceptually unattainable—a condition the authors label constitutive AI unaccountability.

Why does this matter today? Enterprises are racing to commercialize agentic AI while regulators scramble to draft liability frameworks. If a deployment sits inside a constitutive unaccountability configuration, any subsequent legal or compliance effort will be chasing a phantom, leaving companies exposed to reputational damage, regulatory penalties, and unintended societal harm.

What the Researchers Propose

The authors present a three‑stage qualitative study that culminates in a diagnostic framework composed of 20 targeted questions. The framework is organized into three inter‑related clusters:

1. Structural Cluster: examines institutional arrangements, governance hierarchies, and ownership models that shape who can be held liable.
2. Technological Cluster: focuses on system architecture, autonomy levels, and the opacity of learning mechanisms.
3. Normative Cluster: looks at ethical standards, societal expectations, and the presence (or absence) of normative scaffolding.

Each cluster contains multiple categories (nine in total) and 20 thematic indicators that together capture the conditions under which accountability becomes constitutive rather than merely procedural. The framework is operationalized as a questionnaire that can be applied to any AI deployment to surface hidden accountability voids.

How It Works in Practice

Applying the framework follows a straightforward workflow:

1. Scope Definition: Identify the AI system under review (e.g., an autonomous trading bot, a customer‑service agent, or an open‑source research tool).
2. Stakeholder Mapping: List all actors—developers, operators, platform providers, end‑users, and regulators—who interact with the system.
3. Questionnaire Execution: Run through the 20 diagnostic questions, documenting evidence for each indicator.
4. Cluster Scoring: Aggregate answers within each cluster to reveal concentration of unaccountability signals.
5. Interdependency Analysis: Examine the eight directed interdependencies the authors identified (e.g., how a technological opacity can reinforce a structural power asymmetry).
6. Remediation Blueprint: Generate a set of concrete design or governance changes aimed at breaking the identified loops.

What distinguishes this approach from prior checklists is its focus on constitutive properties—features that are baked into the system’s sociotechnical fabric—rather than on superficial compliance artifacts. For example, the framework flags an “inverted anthropomorphism” pattern where the AI agent is the only visible actor, effectively erasing human decision‑makers from the accountability chain.

Evaluation & Results

The researchers validated the framework through three complementary methods:

• Concept‑Centric Literature Review: Mapping 150+ scholarly works onto the nine categories to ensure theoretical coverage.
• Secondary Analysis of 27 Expert Interviews: Coding insights from AI engineers, legal scholars, and sociotechnical researchers to surface real‑world pain points.
• Case Study Application to OpenClaw: Applying the 20‑question instrument to the open‑source agentic system OpenClaw (which includes the Clawdbot and MoltBot agents).

When the questionnaire was run on OpenClaw, 17 of the 20 constitutive unaccountability conditions were detected. Notably, the system exhibited:

• A structural concentration of decision‑making within a single open‑source maintainer community, leaving no clear corporate liability.
• Technological opacity due to self‑modifying code that updates its own policy network without external audit trails.
• A normative vacuum where the community’s informal code‑of‑conduct does not address downstream harms.

These findings demonstrate that the framework can surface deep‑seated accountability gaps that would be invisible to standard risk assessments. Moreover, the interdependency analysis revealed that addressing a single technological opacity without reshaping the governance structure would have limited impact, underscoring the need for holistic remediation.

Why This Matters for AI Systems and Agents

For AI practitioners, the framework offers a pragmatic lens to audit their deployments before they go live. It helps answer critical questions such as:

• Who can be legally pursued if the agent causes harm?
• Do system logs provide enough granularity to reconstruct decision pathways?
• Are there societal expectations that the current design violates?

By surfacing constitutive unaccountability, teams can redesign architectures—e.g., introducing human‑in‑the‑loop checkpoints, modularizing learning components for better traceability, or formalizing governance contracts with clear liability clauses. This proactive stance reduces the likelihood of costly post‑mortems and aligns product roadmaps with emerging regulatory expectations.

From a product‑management perspective, the diagnostic instrument can be embedded into the Workflow automation studio to trigger accountability checks at each stage of the CI/CD pipeline. For enterprises seeking to scale AI responsibly, the framework dovetails with the Enterprise AI platform by UBOS, enabling systematic risk dashboards that surface unaccountability signals in real time.

What Comes Next

While the study makes a strong conceptual contribution, several limitations remain:

• Scope of Empirical Validation: The case study focused on a single open‑source system. Broader validation across regulated sectors (finance, healthcare) is needed.
• Quantitative Metrics: The current framework is qualitative; future work could develop scoring rubrics that translate into risk scores.
• Tooling Integration: Automating the questionnaire through AI‑assisted document analysis would lower the barrier for large‑scale adoption.

Future research directions include:

1. Extending the diagnostic to multi‑agent ecosystems where accountability may be distributed across dozens of interacting bots.
2. Linking the framework with emerging standards such as ISO/IEC 42001 (AI governance) to create a compliance‑ready toolkit.
3. Exploring “accountability‑by‑design” patterns that embed traceability and human oversight directly into model training pipelines.

Practitioners interested in operationalizing these insights can start by piloting the questionnaire on a low‑risk internal prototype, then scaling the process through the UBOS platform overview. By doing so, organizations not only mitigate legal exposure but also build trust with customers and regulators who increasingly demand transparent AI behavior.

Ready to embed accountability checks into your AI development workflow? Explore the UBOS templates for quick start and see how the diagnostic framework can become a living part of your governance toolkit.

Andrii Bidochko

CTO UBOS

Andrii Bidochko is an AI entrepreneur and researcher focused on AI agents, reinforcement learning, and autonomous systems. He writes about the technologies shaping the future of machine intelligence, from frontier models and agent architectures to real-world AI applications.