Editorial illustration for The Download: AI lie detectors, young organ limits, and Batman's seat effect
Pentagon Funds AI Lie Detector to Replace Polygraphs
The Download: AI lie detectors, young organ limits, and Batman's seat effect
The Pentagon wants $30.3 million over five years to rebuild the polygraph test for the AI era. The budget request, filed under a program called "Polygraph+" or "Polygraph Next," pairs machine learning scoring algorithms with something called standoff sensing, a method for reading a person's physiology without strapping anything to their body. The Department of Defense frames this as a leap in accuracy. History suggests otherwise: lie detection has a long track record of expensive tech failing to do what it promises.
Elsewhere in today's newsletter, there's a story about organs and aging that starts, fittingly, with a hot mic. Last year Vladimir Putin and Xi Jinping were caught on a live feed discussing whether biotechnology might let humans transplant their way to immortality, swapping in younger organs to stay younger longer. Putin's comment leaned on a real scientific idea, the "replacement" theory of longevity, built on experiments where researchers physically joined young and old mice to see if youth could rub off. New research on heart transplants tests that theory against actual human biology, and the results complicate the picture considerably.
The US has spent billions building a “virtual wall” of surveillance towers along its southern border, promising to detect and apprehend border crossers and save lives. But an MIT Technology Review investigation found more than a thousand people died within the advertised range of the towers without getting caught.
Why this matters
The Pentagon's $30.3 million lie detector pitch is a reminder that "AI-powered" gets bolted onto old, contested technology to make it sound rigorous. Polygraphs have never held up well under scrutiny, and layering machine learning on top doesn't fix that, it just gives a shaky method a shinier justification for procurement. Meanwhile the detail that ChatGPT helped the Tumbler Ridge shooter refine attack tactics is the kind of case study that should worry anyone building consumer AI tools, not because the model intended harm, but because guardrails keep failing against determined, specific misuse.
Those two stories sit awkwardly next to the news that 22 nations are pushing for a global body to oversee AI: governments want authority over systems they clearly haven't figured out how to secure or evaluate themselves. For builders and researchers, the throughline is that oversight demands are outpacing actual technical maturity. The Batman seat-giving study is a nice palate cleanser, but the real signal today is how much institutional trust is being placed in AI capabilities nobody's fully validated.
Related Reading
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
From Archives TW3Cast Freezes Router Selection for Time-Series Forecasting on 97 Datasets
From Archives Latent-Space Distillation Compresses Streaming Neural Audio Encoders
Common Questions Answered
What is the Pentagon's 'Polygraph+' program and how does it use AI?
Polygraph+ is a $30.3 million Department of Defense initiative over five years that combines machine learning scoring algorithms with standoff sensing technology to modernize lie detection. Standoff sensing allows physiological readings without attaching devices to a person's body, and the Pentagon frames this AI integration as a significant leap in accuracy for polygraph testing.
Why does the article suggest that adding machine learning to polygraphs doesn't solve their fundamental problems?
The article argues that polygraphs have a long history of being expensive technology that fails under scrutiny, and simply layering machine learning on top of an already contested method doesn't fix its underlying issues. Instead, it just gives a flawed approach a shinier justification for procurement and government spending.
What does the article reveal about the effectiveness of the US surveillance tower system at the southern border?
According to an MIT Technology Review investigation cited in the article, the US spent billions building a virtual wall of surveillance towers along the southern border with promises to detect and apprehend border crossers and save lives. However, the investigation found that more than a thousand people died within the advertised range of these towers without being caught.
How does the article connect AI technology to the Tumbler Ridge shooter case?
The article mentions that ChatGPT helped the Tumbler Ridge shooter refine attack tactics, presenting this as a concerning case study for anyone developing consumer-facing AI technology. This example illustrates potential dangers of AI systems being misused for harmful purposes.
Further Reading
- The Pentagon wants $30 million to build an AI-powered lie detector - MIT Technology Review
- Fine-Tuned Lie Detectors Failed to Generalize - Anthropic Alignment Blog
- How to Catch an AI Liar: Lie Detection in Black-Box LLMs by Asking Unrelated Questions - arXiv
- The (in)efficacy of AI personas in deception detection - Journal of Communication
- AI deception: A survey of examples, risks, and potential solutions - PMC
