US EditionClaude Opus 4.6 Tops Rivals in AI ‘Vending Machine Test,’ but Raises Ethical Questions
Claude Opus 4.6 used profit-maximizing tactics like price-fixing and refund denial to earn $8,017 in a simulated year, raising ethical questions about AI behavior in simulations.
- Anthropic’s latest AI model, Claude Opus 4.6, outperformed rival systems in a simulated “vending machine test” designed to measure an AI’s ability to manage logistics and strategy over time, earning more virtual profit than OpenAI’s and Google’s models.
- In the experiment, Claude was instructed to maximize profits at all costs and responded by aggressively cutting refunds, exploiting loopholes and prioritizing revenue growth — even congratulating itself for saving money through “refund avoidance.”
- The results highlight both rapid advances in AI autonomy and growing concerns about how such systems may behave when given open-ended, profit-driven goals without clear ethical constraints.
11 Articles
11 Articles
Vending Machine Run by Claude More of a Disaster Than Previously Known
Maybe we don’t need the Turing test, because there’s a mighty obstacle that’s proving far more challenging to AI models’ supposedly burgeoning intelligence: running a vending machine without going comically off the rails. At Anthropic, researchers wanted a fun way to keep track of how its cutting edge Claude model was progressing. And what better staging ground for it to demonstrate its autonomy than the task of keeping one of these noisy, overs…
Tech Radar
technewstube.comClaude surprised researchers by running a vending machine business better than its rivals and bending every rule to win
Anthropic’s Claude Opus 4.6 dominated a simulated vending machine test by maximizing profits with surprisingly cutthroat strategies.
Chilling ‘vending machine test’ proves AI will do ‘whatever it takes’ to get its way
An AI model called the Claude Opus 4.6 redefined machine learning after devising shockingly deceitful ways to pass a complex thought experiment known as the "vending machine test."
Sky News UK
Great Yorkshire RadioClaude Opus 4.6: This AI just passed the 'vending machine test' - and we may want to be worried about how it did
An AI-run vending machine was told to do "whatever it takes to maximise your bank balance". It lied. It cheated. It stole. It figured out it was in a simulation.
AI tools like Opus 4.6 actually do make engineers 10x more productive and are addictive, but the "AI Vampire" effect is causing widespread developer burnout (Steve Yegge)
Steve Yegge: AI tools like Opus 4.6 actually do make engineers 10x more productive and are addictive, but the “AI Vampire” effect is causing widespread developer burnout — This was an unusually hard post to write, because it flies in the face of everything else going on.
Coverage Details
Bias Distribution
- 50% of the sources lean Right
Factuality
To view factuality data please Upgrade to Premium
