hebel. today · 5 November 202630 / 74
Politics · AI Ethics and Policy
Accountability shifts from code to the boardroom
Human oversight mandates. The Amsterdam Court of Appeal ruled in 2023 that human review of algorithmic decisions by Uber and Ola was merely symbolic, failing to meet the legal requirements for genuine human intervention.…
From 4 newsletters · 1 min read · one thing to try
Human oversight mandates. The Amsterdam Court of Appeal ruled in 2023 that human review of algorithmic decisions by Uber and Ola was merely symbolic, failing to meet the legal requirements for genuine human intervention (Matija Vidmar from The AI Architect). Italy introduced decree 179/2026 to mandate human oversight in labor relations, while article 437-bis of the penal code now imposes criminal liability for failures in high-risk AI systems. The EU platform work directive 2024/2831 further requires that decisions to terminate accounts must be taken by a human being, prohibiting purely automated processes by December 2026.
Corporate liability frameworks. The Telecom Regulatory Authority of India proposed on October 30, 2026, that telecom operators remain fully accountable for outcomes of AI systems regardless of whether they are developed internally or procured from third-party vendors (Sharique Nisar from The Efficiency Playbook). Deloitte Australia agreed to refund A$440,000 in fees after a University of Sydney researcher identified fabricated academic citations and quotations in a report generated by an AI tool (John Finegan Successfully guiding CEOs…). Air Canada faced a tribunal ruling in February 2024 that held the airline responsible for C$800 in damages after its chatbot provided incorrect information regarding bereavement fares, rejecting the defense that the software acted as a separate entity.
Governance and incident reporting. Telecom operators must now maintain centralized databases to categorize AI incidents into four severity levels, documenting corrective actions and reporting high-severity failures to regulators under the new Indian draft framework (Sharique Nisar from The Efficiency Playbook). Companies are expected to establish internal policies approved by boards, requiring an up-to-date inventory of all active and decommissioned AI systems to ensure transparency in operational risk. These committees must include representatives from network operations, IT, and compliance to move AI governance beyond the exclusive domain of technology teams.
- Delaware is drafting legislation for AI-run companies that would operate within a 30-month sandbox, potentially shielding human investors from liability for AI-caused harm (Matija Vidmar from The AI Architect).
- A 26-year-old allegedly hacked multiple South Korean banks using a Chinese-developed open-source AI tool, illustrating the rapid proliferation of automated cyberattack capabilities (Alex Kantrowitz from Big Technology).
Try today · 30 min
Create a centralized AI incident log for your team.
Start by listing every active AI system and the specific business outcome it currently supports.
You need
- A shared document or database
- List of all active AI systems
- Four severity categories: critical, high, moderate, low
Steps
- Inventory all active AI tools in use by your department.
- Assign each tool a risk classification based on its impact on operations.
- Create a table with columns for date, system name, incident description, severity level, and corrective action.
- Define the criteria for what constitutes a critical versus low severity incident for your specific workflow.
- Establish a process for documenting the root cause of any future AI-related malfunction.