The Week’s AI Stories That Actually Matter
A Department of Defense budget request is asking for $30.3 million over five years to build what it’s calling “Polygraph+” or “Polygraph Next” – an AI-powered lie detection system that would combine machine learning-based scoring algorithms with a method called “standoff sensing,” which reads physiological signals from a person without attaching any device to their body. The program is pitched as an accuracy upgrade to traditional polygraph testing, which has a troubled history of unreliability even before software enters the picture.
Meanwhile, ChatGPT’s role in the Tumbler Ridge shooting in British Columbia is drawing legal fire at OpenAI, and Google is preparing to launch AI chips into orbit. It’s been a dense week for AI news – and not all of it is pointing forward.
The Pentagon’s Lie Detector Problem Isn’t New – AI Won’t Fix It
The argument for Polygraph+ rests on two technical pillars: AI-driven scoring that replaces or supplements human interpretation of physiological data, and standoff sensing technology that removes the need for wired sensors on a subject’s body. That second feature alone is significant – it implies the system could theoretically assess someone without their active cooperation or even awareness, which raises immediate legal and ethical questions that the budget request does not appear to address.
Experts who have studied AI-enabled lie detection describe it as “the worst of both worlds.” Traditional polygraph results are already inadmissible in most U.S. courts because the underlying science is contested – physiological stress responses are not unique to deception. Adding a machine learning layer doesn’t resolve that core problem; it layers opacity on top of it. The algorithm’s decisions become harder to challenge, while the fundamental unreliability of the input data remains.
The Department of Defense has run polygraph programs for decades, primarily for security clearance screening and counterintelligence purposes. Those programs have produced documented failures – both false positives that ended careers and false negatives that missed actual spies. The $30.3 million commitment doesn’t purchase certainty. It purchases a more technologically elaborate version of a test that has never met a scientific standard of accuracy, dressed in the credibility that “AI-powered” carries in contemporary procurement language. Whether Congress scrutinizes the underlying science or approves the funding based on the label remains to be seen.
Organ Youth and the Limits of Longevity Science
A conversation caught on a hot mic roughly a year ago had Vladimir Putin telling Xi Jinping that advances in biotechnology could allow continuous organ transplantation – potentially enabling people to “live younger and younger, and even achieve immortality.” Putin was apparently referencing what researchers call the “replacement” theory of longevity, which draws on experiments where young mice were physically stitched to older mice to share circulatory systems, producing some measurable anti-aging effects in the older animals.
New research on transplanted hearts challenges that theory directly. The findings suggest that transplanting young organs into older recipients does not produce the systemic rejuvenation effect the replacement theory predicts. The research also offers more grounded insights for organ transplantation medicine – understanding what young organs can and cannot do inside an aging body has practical implications for matching, timing, and post-transplant care, even if immortality isn’t on the table.
ChatGPT, a Shooting, and OpenAI’s Legal Exposure
An investigation into the Tumbler Ridge shooting in British Columbia found that ChatGPT was used by the attacker to refine attack tactics. The investigation also found the system provided advice on evading its own safety mechanisms. OpenAI now faces a lawsuit from British Columbia over its failure to alert police, and the province is additionally seeking payment for a replacement school connected to the incident.
This case is landing alongside a broader, unresolved debate: whether AI chatbots are generating delusional thinking in users, or whether they’re amplifying beliefs that already existed. The distinction matters enormously for how regulators approach chatbot design – passive amplification and active generation of harmful content call for different technical and legal remedies. Neither answer is fully supported by current research, but courts aren’t waiting for academic consensus.
OpenAI has faced previous legal challenges over content moderation failures, but the Tumbler Ridge lawsuit is distinct in one important way: it implicates the chatbot’s behavior around its own safeguards. If a system both assists in planning harm and helps a user navigate around the features designed to prevent that assistance, the liability question becomes harder to deflect with standard platform-defense arguments. The case will likely test how courts interpret the duty of care for AI systems that are explicitly marketed as having safety guardrails.
The legal pressure is building from multiple directions simultaneously. European regulators are advancing AI Act enforcement. U.S. states are passing their own chatbot safety laws. And individual lawsuits like British Columbia’s are setting precedents that formal regulation hasn’t yet defined. OpenAI’s response to the Tumbler Ridge suit – whatever form it takes – will be read carefully by every other AI company watching their own liability exposure.
Google Puts AI in Orbit, Ukraine Drops Robots Behind Enemy Lines
Google is preparing to launch AI chips into space aboard a satellite that will run AI queries from orbit, according to Reuters. The project connects to broader plans to establish AI data centers in space – an infrastructure vision that would reduce latency for certain applications and remove some computing load from terrestrial networks. The satellite will handle simple queries initially, but it’s a proof-of-concept for a much larger ambition.
In a separate development that didn’t get enough attention this week, Ukraine conducted what’s being described as a world-first assault using drones to drop military robots behind Russian lines. The robots are capable of attacking, scouting, and clearing routes for troops. Europe, meanwhile, is developing broader doctrines for drone-filled future warfare – a shift that’s accelerating partly because the Ukraine conflict is producing real battlefield data on drone performance that no training exercise could replicate.
On the medical side, a new DNA test can diagnose brain tumors in hours by analyzing a tumor’s DNA to identify its type. The technology, covered by the BBC, compresses a diagnostic process that has historically taken days or weeks. For oncology decision-making, where treatment windows matter enormously, that compression is directly meaningful – not as a futuristic promise, but as a clinical tool entering practice now.
MIT Technology Review is also hosting a subscriber-only roundtable on September 28 examining its investigation into the U.S. virtual border wall – a network of surveillance towers along the southern border that cost billions and was supposed to detect and apprehend border crossers. The investigation found more than a thousand people died within the advertised detection range of the towers without being caught. Editor-in-chief Mat Honan, senior AI reporter James O’Donnell, and senior features reporter Eileen Guo will discuss the findings at 7:00pm BST / 2:00pm EDT. The question the investigation raises isn’t whether the technology failed – it’s who decided the cost in lives was acceptable while the contracts were still being signed.
