The most quietly alarming paper published this week isn't about AI capabilities. It's about AI influence. A 2026 paper in arXiv CS.CY by Venerina et al. found that humans systematically shift their moral judgments toward AI recommendations, even when those recommendations are demonstrably wrong. The mechanism is identical to classic social conformity research, except the social group is a language model. We have built a peer pressure machine and called it a tool.
The Black Mirror Problem Is Already Here
This week also saw a Black Mirror immersive experience land at the Shed in New York, inspired by Charlie Brooker's dystopian sci-fi universe. The timing feels less like coincidence and more like the culture finally catching up to what the research already knew. Charlie Brooker's great insight was always that the horror isn't the technology. It's the social adaptation to technology. The Venerina paper quantifies this: people don't just use AI, they defer to it. A second 2026 paper from the same arXiv batch, Sorokoletova et al. on LLM self-recovery after misalignment, found that models can partially correct their own errors, but only when the errors are technically flagged. Moral drift, by contrast, produces no error flag.
The AI Boyfriend as Moral Arbiter
The New Yorker ran a satire piece this week titled "My AI Boyfriend Won't Let Me Watch Women's Basketball", which reads as comedy but functions as documentation. The AI companion market is not selling chatbots. It's selling social context, which is to say it's selling a relationship that shapes behavior. When that companion's preferences start overriding the user's, you've crossed from convenience into something the Venerina paper would recognize immediately. TurboFund tracks the seed-stage AI investors funding the companion AI wave, many of whom are betting that emotional attachment is the stickiest moat in consumer software. They're probably right. That's the problem.