Who is Accountable? Is it AI?

November 17th, 2026 by Camille Bexley

How can you trust the very AI you are using? 

Two events, 3 months apart and we are still asking who owns the accountability when AI goes wrong. 

This begs the question, what is missing? 

Independent AI Assurance Lead​ing the Way

I am still buzzing from today’s Humans in AI Leadership Think Tank: Who Owns the Consequence? Unlike so many industry panels drowned in buzzwords, this session focused squarely on stamping out AI hype. Hosted by Aquent, guided by Fiona Wilhelm’s sharp facilitation and framework, the session placed a refreshing emphasis on human intelligence and critical thinking as the ultimate gatekeepers of artificial intelligence innovation. The room brought together a diverse mix of senior AI leaders and practitioners. In the spirit of the Chatham House Rule, I won’t attribute specific insights to individuals, organisations, or sectors, as the true value of these forums lies in the candour they protect. 

Roundtable discussion (stock image)

However, 3 months ago I watched a panel including our CEO Founder Adam Sandman together with NSW Services and Pickles Auction at the Sydney DevOps Summit. I want to share my own reflections on both events with the core question still being explored: Who actually owns the consequence? Because the timing couldn't be more poignant. Over the past week alone, we witnessed SpaceX catch rockets out of mid-air, major email and banking outages, and an entire country's fleet of planes grounded.

Our technological capabilities are soaring, yet our underlying systems remain deeply fragile.

Across the APAC region and globally, I see organizations struggling to bridge the gap between experimental AI and enterprise-grade reliability. Here is where traditional AI governance is falling short, and how we can build true accountability into our everyday.

1. Data Integrity Is Still ​a Core Matter

When I was marketing Analytics for SaS, we advised clients: ​"Garbage in, garbage out". It still is the case today. While AI relies entirely on the data we feed it, organisations rarely stop to ask how accurate and reliable that underlying data truly is. No matter how advanced the model, poor data integrity inevitably leads to flawed and skewed outputs. Adding my point 4 and 5 below, we need to challenge the assumption of data completeness. 

The Solution: Before trusting AI to make decisions, organisations must validate how their systems handle data edge cases, hallucinations, and potential data leakage. ​The Inflectra Platform ​allow teams to stress-test AI agents against adversarial inputs and real-world data scenarios before they ever reach production, ensuring systems do not confidently deliver the wrong answers.​ Inflectra.ai helps the data enrich and get smarter over time learning from each run, injestion and result.

 

2. Independent AI Testing of AI

Traditionally, organisational governance focused across four typical foundational areas:

  • People Layer: Requirements, user story input, and final human gatekeeping.
  • Process Layer: How work moves from idea to execution.
  • Policy Layer: The rules, ethics, and compliance standards governing the business.
  • Production Layer: SDLC supported by a traditional "Policing Layer" (standard testing).

I am calling for a Fifth Layer: independent AI testing AI for red flags. Rather than waiting until a final policing stage just before release, or worse, after deployment, organisations need continuous Quality Assurance (QA) with humans firmly in the loop at every step of progress. With the sheer scale and volume AI has brought us, teams are experiencing AI burnout from moderating AI Generated Artifacts and sometimes AI Slop. It is simply not possible to manually moderate everything all the time.

The Solution: Static test scripts and manual checklists cannot keep up with dynamic AI agents. SureWire.ai, together with SpiraPlan, operationalizes this fifth layer by using specialized testing and judge agents to proactively probe your AI. It acts as a continuous pulse check—catching behavioral drift, prompt injections, and policy violations early while keeping human reviewers in control of the final verdict. Meanwhile, SpiraPlan’s AI enriches bidirectional requirements and risk management, not just offering advice, but executing tasks on your behalf with your explicit approval. Think of SureWire as an ongoing pulse check or a dipstick test - at scale: flagging anomalies or deviations from core requirements in real time.

 

3. End-to-End Traceability

One of the biggest governance blind spots today is defining responsibility​ and accountability when AI goes wrong. Is it the employee who prompted it, the manager who signed off, the tech leaders who advocated purchase, the CISO, or the Board? Or does liability sit with the LLM provider, the application vendor, or the implementation partner? 
When an incident occurs, the glaring missing piece in most enterprises is end-to-end traceability. Without a clear audit trail, accountability quickly turns into a finger-pointing exercise, especially when enterprise projects involve cross-functional cross-border teams operating under different legislation, offshoring arrangements, and varied regional policies.​ Then add the complexity of vibe coding and ai assistance, hybrid teams of human and ai. 

The Solution: When combined with governance frameworks like Inflectra’s SpiraPlan, SureWire.ai creates a closed-loop assurance system. SureWire’s "Key Findings" document exactly which scenarios were tested, what behavior was observed, and why a result passed or failed. This gives leaders, CISOs, and boards clear, audit-ready evidence linking initial requirements to final deployment.​ Inflectra's platform auto-tags what ai generates or augments as well as human validation every step of the SDLC. 

 

4. Shared Context 

When we look at the intersection of intellectual property, organisational blind spots, and shared context, a troubling pattern emerges. Often, an AI doesn't go rogue. It does exactly what it was told to do, yet still creates massive reputational or operational harm. Why? Because it lacked shared specific context and a broader "perspective" of ​possible impacts/consequences.

The Solution: Building effective guardrails against "unintended" consequences  requires testing AI across multi-turn conversations and complex workflows to see how it behaves when pushed off-script or even on brief under different circumstances. SureWire.ai helps uncover these contextual blind spots by evaluating AI behavior against policy-driven scenarios, ensuring your guardrails hold up under nuanced, real-world pressure.

 

5. Protecting Irreplaceable Tacit Knowledge

While there sure is a fear of AI replacing our work or even us as a species if left unattended. Today reinforced one undeniable truth: tacit knowledge is irreplaceable. Rooted deeply in human experience, intuition, and practical wisdom, tacit knowledge is notoriously difficult to write down or codify into a prompt. As the saying goes, coined by Hungarian-British polymath Michael Polanyi in his 1966 book: "we can know more than we can tell". It is the subtle art of navigating stakeholder dynamics, or cooking an extraordinary meal through intuition and improvisation rather than strictly following a recipe/user manual. It is practical, personal, and inherently human.

While AI can learn and get smarter. AI can accelerate our tasks and workflows, significantly expand our cognitive repertoire to enhance our very human ways of working and our existence. And platforms like SureWire.ai and SpiraPlan can give us the confidence that our AI agents are safe, reliable, and compliant. But at the end of the day, AI is a tool. Human intelligence, lived experience, and critical thinking must remain at the helm. We must ensure we always retain the choice by Assuring the very AI we use. 


About the Author

Camille Bexley

Camille Baumann is the Regional Director APAC at Inflectra. In this role, she's responsible for Sales, Solutions, Customer Success, and Alliances across the region. At Inflectra, Camille combines her deep expertise in digital transformation with a passion for customer-centric strategy, helping organizations adopt robust software quality assurance and lifecycle management solutions—powered by Inflectra’s SpiraPlan, Rapise, and related technologies.

Spira Helps You Deliver Quality Software, Faster and with Lower Risk.

Get Started with Spira for Free

And if you have any questions, please email or call us at +1 (202) 558-6885