AI-Assisted Surgical Decision-Making Without Contemporaneous Decision Records

Why this document exists

This use case is published through the AI OSI™ initiative to examine a publicly reported incident involving AI-assisted surgical systems.

It does not assess clinical correctness, assign liability, or offer regulatory judgment. Its purpose is narrower: to illustrate how harm involving AI can be compounded by the absence of contemporaneous, human-readable governance records, and how that absence is neither necessary nor inevitable.

The failure described here is not a failure of artificial intelligence. It is a failure of institutional memory.

The situation as it now exists

AI systems are increasingly embedded in surgical and clinical environments. They assist with visualization, navigation, anatomical identification, monitoring, and risk assessment.

These systems are rarely introduced through a single, explicit governance moment. Instead, they are adopted through procurement pathways, vendor training, and gradual normalization within clinical practice.

Over time, reliance increases. Scrutiny decreases. Authority becomes implicit. When AI influences care, it often does so quietly—shaping attention and judgment without leaving a durable record of how it was used, how influential it was, or who authorized that reliance.

When something goes wrong, institutions are left reconstructing decisions that should have been recorded at the moment they occurred.

What the public reporting revealed

According to public reporting, AI-assisted systems were used in surgical contexts where misidentification or incorrect guidance occurred and patient harm resulted.

Afterward, it was not possible to clearly determine:

• how influential the AI system had been in the decision process

• whether its use exceeded its intended or approved role

• who explicitly authorized reliance on it at the point of care

• what constraints or limitations were understood at the time of use

Statements following the incident emphasized complexity, training issues, or distributed responsibility. What was absent was a contemporaneous governance record capturing how reliance on AI was authorized when it mattered.

That absence is the central issue.

What actually failed

The failure was not model accuracy alone, system design alone, or clinician competence alone.

The failure was the absence of a simple, durable governance record capturing:

• who decided to rely on an AI system

• under what conditions that reliance was permitted

• what risks were understood at the time

• and until when that authorization remained valid

Without such records, accountability collapses into retrospective reconstruction. Reconstruction introduces hindsight bias. Hindsight bias invites adversarial reinterpretation. Institutional memory erodes precisely when it is most needed.

This is not merely a documentation gap. It is a governance failure in evidentiary continuity.

How the AI OSI™ framework would have changed what could be known

The AI OSI™ framework does not prevent error. It preserves evidentiary continuity around error.

If a standing governance record had existed under the AI OSI™ framework defining how the system was approved for use, what it was not approved to do, and who held responsibility for that authorization, then scope drift would have been visible before harm occurred.

If a contemporaneous decision record had been created at the moment of care, it would have captured:

• the degree of AI influence in the decision

• whether human override was substantive or nominal

• and what risks were understood at the time of reliance

That record would have preserved the decision as it existed before outcomes reshaped interpretation.

If a first-capture incident record had been created immediately after the unexpected outcome, it would have preserved operational reality before legal framing, institutional messaging, or vendor alignment altered the narrative surface.

None of this requires advanced technical capability. It requires naming responsibility at the moment it exists.

Why this was not inevitable

Nothing about this incident required new models, additional regulatory authority, or prohibition of AI in clinical environments.

It required a small number of contemporaneous governance records, created at the moment decisions were made, by identifiable human owners, documenting AI’s role in consequential decisions.

The absence of those records did not cause the clinical harm directly. It caused the inability to clearly reconstruct, defend, or learn from the conditions under which the harm occurred.

What this illustrates more broadly

This failure mode is not unique to healthcare.

It appears wherever AI systems influence decisions affecting:

• physical safety

• employment

• financial access

• legal rights

• or institutional outcomes

Across these domains, the same structural question emerges:

Who decided to trust the system, and where is that decision recorded?

When that question cannot be answered cleanly, accountability becomes interpretive rather than evidentiary.

Conclusion

AI did not make this situation inherently unknowable. The absence of contemporaneous governance records did.

The AI OSI™ initiative exists to replace that absence with structured institutional memory—created at the moment decisions occur, not reconstructed after outcomes are known.

This is not a claim of safety or certainty. It is a commitment to evidentiary continuity.

It is a refusal to forget decisions that cannot be undone.

References

Reuters, AI enters operating room; reports arise of botched surgeries, misidentified body parts, February 9, 2026.
https://www.reuters.com/investigations/ai-enters-operating-room-reports-arise-botched-surgeries-misidentified-body-2026-02-09/