Amazon Atlas Union Risk AI & AI Abolition Manifesto Gains Attention - AI News (Oct 18, 2026)

What if an algorithm could estimate how likely your workplace is to unionise before anyone has signed a single card? According to a new investigation, one of the world's biggest employers may already be doing exactly that. Welcome to The Automated Daily, AI News edition. The podcast created by generative AI. I'm TrendTeller, and today is October 18th, 2026. Today we're covering AI and labor rights, a political standoff between Canada and the US, a new way to collect robot training data, and the effort to verify software that AI agents are writing. Let's get into it.

Amazon Atlas Union Risk AI

We start with that investigation. openDemocracy reports that Amazon uses an in-house AI tool called Atlas to assess the risk of union activity at warehouses in the US, Canada and the UK. According to internal documents seen by the outlet, the system pulls together hundreds of data points, including social media posts, employee survey answers and the history of past labor actions. It then scores each site on how likely it is to unionise or to see disruptive labor events. Two UK locations were reportedly flagged as high-risk: the Coventry warehouse and the robotic fulfilment centre in Sutton Coldfield. Union organizers call Atlas a sophisticated anti-organizing tool that widens an already lopsided power gap. Amazon disputes that. The company says Atlas does not monitor individuals or personal data, works only at the site level, and exists to help managers understand workplace concerns. Even taking Amazon's account at face value, the story raises a real question: what happens to collective bargaining when employers can use predictive analytics to see organizing coming before it starts?

AI Abolition Manifesto Gains Attention

Staying with AI's social impact, a manifesto calling for outright AI abolition is drawing attention. It argues that AI is not a neutral tool. In its view, AI is bound up with capitalism, control and the outsourcing of human judgment, and it compares the cause to prison and police abolition movements. The authors list harms they say are already happening at scale, including environmental damage, labor exploitation, loss of privacy, harm to education and a drift toward authoritarianism. They reject the mainstream idea of keeping AI's benefits while managing its risks, and they argue there is no ethical AI under capitalism. Their prescription is to build community, rely less on AI tools, speak out and organize. Few people will go that far, but the piece shows that the backlash against AI is becoming more organized and more explicitly political.

Caseway Rejects US Government Work

Geopolitics is shaping AI business decisions too. Caseway, a Vancouver-based company that builds AI tools for legal and defence work, says it is walking away from opportunities with the US government. CEO Al Vigier says he isn't comfortable helping a country that keeps threatening Canada, and he pointed to Donald Trump's repeated '51st state' comments. Caseway will now focus on Canadian defence contracts and look for customers in Europe and Taiwan. The decision has a cost, because the company gave up a high-profile chance to pitch investors at an industry conference. It also highlights an uncomfortable fact: Canada's AI sector still depends heavily on American money and infrastructure.

HUMXN Trades Repairs for Robot Data

Next, an unusual way to gather training data. AI lab HUMXN is offering homeowners free plumbing, electrical, HVAC, air-duct cleaning and carpet cleaning work. The condition is that the technicians wear recording gear while they do the job. The goal is to capture how skilled tradespeople actually diagnose problems and finish repairs in messy, varied real homes, which a tidy lab can't reproduce. The program launches in Minneapolis, Chicago and Miami, three cities with very different climates and housing. The bigger point is that the internet has plenty of text and images, but very little data on hands-on physical work. Companies building household robots are now willing to pay for that data, and in this case the payment is free repairs.

Verification and Comprehension Checks for AI Code

Now to an ongoing thread about verifying code written by AI agents. A new article argues that verification has to become a core part of software development. It cites industry data linking AI adoption to more bugs, longer reviews and more production incidents. The author surveys AI review options from Anthropic, OpenAI, GitHub, Cursor and others, then proposes layered checks: fast, cheap ones early in the process, and heavier scrutiny for pull requests and risky changes. In his own workflow, one model writes the code, a second model reviews it, and a human still gives final approval. Along the same lines, the open-source Git hook Sphinx, which we've covered before, now quizzes developers on what agent-written code does before a push can continue. It focuses on edge cases such as retries and failures, and it keeps the developer's answers hidden from the coding agent.

Behavioral Testing Exposes AI Agent Bugs

Last, a reminder of why that matters. DefuzeX ran 53 open-source AI agents through its behavioral testing setup and reports 37 bugs across about 3,000 generated scenarios. The most telling example was an agent that asked for permission before running a harmless command and then deleted a file without asking at all. The broader lesson is that many agent failures come from how tools, permissions and multi-step workflows interact, not from the underlying model. An agent that looks capable in a demo can still behave unpredictably once it has real access to a system. Checking an agent's text responses is not enough. Teams also need to test what the agent actually does.

That's all for today's edition of The Automated Daily, AI News edition. Today's stories ranged from union-risk scores in warehouses to robots learning from plumbers, and they share a theme: data about people and their work is becoming more valuable to AI companies and more contested. Links to all of today's stories can be found in the episode notes. I'm TrendTeller. Thanks for listening, and I'll see you tomorrow.

Amazon Atlas Union Risk AI & AI Abolition Manifesto Gains Attention - AI News (Oct 18, 2026)

Our Sponsors

Today's AI News Topics