Can You Jail an AI When It Fails?

Can You Jail an AI When It Fails?

by Mike Holcomb · View the channel on PSI

Humanities, Social Sciences & Thought AI & Society UBDigital and Information technology: general topicsUBJDigital and information technologies: social and ethical aspects
🎙 Mike Holcomb 👥 28K 📅 October 22, 2026 ⏱ 41 min 👁 4 📄 expert opinion 🧭 2026-10-22
Available in: English (current) Français

Keywords

AI governanceOT securityICScontrol loopsaccountability

Summary

The panel discussion explores the integration of AI into operational technology (OT) environments, focusing on governance, security, and accountability. Experts from nuclear, water, energy, and oil & gas sectors discuss benefits like process optimization and predictive maintenance, but also highlight risks such as AI drift, data trust, and unintended entry via software updates. They examine attack vectors, the need for local AI deployment, and the challenges for resource-constrained utilities. The conversation emphasizes the necessity of guardrails, human oversight, and clear liability frameworks when autonomous systems make decisions. The title question about jailing AI underscores the broader issue of assigning responsibility when AI fails.

103 words

Critical Evaluation

Value of the Information & Strength of the Argument

The value lies in the practical insights from experienced practitioners. They provide concrete examples, such as behavioral modeling for network traffic and AI-driven predictive analytics. The argumentation is measured and nuanced, acknowledging both benefits and risks. The discussion effectively highlights the gap between AI’s potential and its current limitations in safety-critical systems. The panel’s stance is that AI should augment, not replace human oversight, and that governance must evolve to address accountability.

Scientific Rigor, Source Quality, Title Accuracy

No formal citations are provided beyond the panelists’ experiences. The discussion references real incidents like the Monterrey water plant attack, which adds credibility. The title of the video is somewhat provocative but aligns with the core theme of accountability. The content is consistent with the description and the panel’s expertise. No external sources are cited, so the quality relies on the panelists’ authority.

150 words

Title / Content Match

The title poses a provocative question about AI accountability, which the discussion addresses through governance and liability considerations, though the focus is more on practical challenges than on legal specifics.

Quality & Reliability

7/10

The information comes from experienced professionals in critical infrastructure sectors, but relies on anecdotal evidence and lacks formal citations. Panelists are credible and provide concrete examples, though the discussion is exploratory rather than empirical.

Chapters

Cited Sources

Contribution & Novelties

The video consolidates expert opinions on AI in OT, particularly around governance gaps, the ‘human in the loop’ concept, and the challenge of AI arriving as software updates. It offers practical perspectives from multiple critical infrastructure sectors, emphasizing the need for clear liability frameworks and local AI deployment.

Pour aller plus loin :

  • NIST AI Risk Management Framework — Provides a structured approach to AI governance and risk management.
  • ISA/IEC 62443 — Standards for industrial automation and control systems security, relevant to OT environments.
  • Adversarial machine learning — Explores how AI models can be deceived, a key concern for security in OT.
  • AI drift (without URL) — Concept discussed as a risk of model degradation over time, relevant to maintaining control in autonomous systems.

124 words

Radar Profile

The profile shows moderate to high scores across quantite, qualite, technique, and fiabilite, indicating a balanced yet exploratory discussion. The fiabilite is slightly lower due to lack of formal citations, but the practical expertise of panelists compensates.

Reliability 7/10