When AI Agents Act for Us, Who Answers When They Get It Wrong?

In this Friday special of our AI news podcast, Aiman, Mei Ling and Raj ask who answers when an AI agent sends, spends or books something wrong, and how much we should let agents do. Mei Ling leads a longer, reflective episode that covers four things: the strongest case for agents, what they mean for jobs and trust in Malaysia, what history says about automated mistakes, and each host’s position.

AI-YA Daily is researched, written and voiced with AI, and edited by the AI-YA team.

Also in this episode:

• Anthropic’s amended IPO filing and its “may imminently meet” warning.
• Visa’s agent-payment testing with four Malaysian banks.
• An Exabytes agent that answers a durian seller’s customers through the night.
• Stanford’s latest data on young workers.
• Air Canada’s chatbot ruling, the 2010 flash crash, Knight Capital, Uber’s Tempe crash, the Tesla Autopilot verdict and Australia’s Robodebt.
• What Odd Lots, The Diary of a CTO, The Ezra Klein Show and a Washington briefing said about agent liability, and what they miss about Malaysia.

Timestamps 0:04 Cold open: the airline that blamed its chatbot 0:53 Meet the hosts: when AI acts for us 2:05 The case for AI agents, from Visa’s bank tests to a durian seller 6:22 Jobs, trust and who carries the blame 11:02 History’s warnings: flash crash, Knight Capital, Uber, Robodebt 16:11 A word from Hitower IT 17:00 What four podcasts say about agent liability, and what they miss about Malaysia 21:53 The Friday positions: how much should an agent do without asking? 26:02 One thing to do this weekend 26:39 Tomorrow: week in review

Stories and sources

• Anthropic’s amended S-1 and the “may imminently meet” threshold (TechCrunch)
• Anthropic’s revenue and spending commitments in the amended S-1 (Yahoo Finance)
• Anthropic says rogue AI agents pose uncertain legal risk (Reuters via The Star)
• Gobind on the AI Governance Bill’s scope (The Star)
• Four Malaysian banks join Visa’s AI payment testing programme (RinggitPlus)
• Exabytes founder on agentic AI for Malaysian SMEs (Digital News Asia)
• AWS research on Malaysia’s AI and agentic AI adoption (Tech Wire Asia)
• Canaries in the Coal Mine, revised (Stanford Digital Economy Lab)
• MDEC’s 2025 digital investments and expected jobs (TechNode Global)
• AI Untuk Rakyat modules and free tools (The Star)
• Jensen Huang: IT will be the HR department of AI agents (Fortune)
• Moffatt v Air Canada (McCarthy Tétrault)
• Regulators blame an algorithm for the 2010 flash crash (Computerworld)
• SEC charges Knight Capital under the market access rule (SEC)
• Uber safety driver convicted over the Tempe crash (New Mobility)
• Florida jury orders Tesla to pay in a fatal Autopilot crash (Yahoo Finance)
• Robodebt victims win record settlement (The Canberra Times)
• Moral Crumple Zones, Madeleine Clare Elish (ESTS)
• Replit AI agent wipes a production database (eWeek)
• National AI Action Plan, AI Malaysia and MY-AISafe (GovInsider)
• PDPA amendments and penalties (Rödl & Partner)
• Odd Lots: AI Agents Are Already Changing the Price of Goods
• The Diary of a CTO: Who Pays When an AI Agent Is Wrong?
• The Ezra Klein Show with Jack Clark (transcript)
• Congressional Internet Caucus: AI Rogue Agents, Who’s Liable?

Standing thread: The big questions (Friday special).

AI-YA runs AI lectures, workshops and implementation training for Malaysian companies. Details at ai-ya.my/podcast.