A skin patch tested on about 2,000 people in India, South Africa and Latvia detected active tuberculosis with approximately 90% accuracy, according to Technion scientist Hossam Haick. It costs a little over a dollar to manufacture and works for months, he said. But real-world tests still need to show whether it can reliably identify disease. In health technology, promise can outrun proof.
The apparatus behind the patch
Haick’s team developed the patch with funding from the Bill & Melinda Gates Foundation. A four-centimeter-by-four-centimeter version contains six sensors that pick up substances released by skin and send data to a phone app. Laboratory engineer Walaa Saliba said the system compares readings with statistical databases of people with and without diseases, then alerts users if readings shift toward illness. Haick expects commercial use within two or three years, initially for a small number of diseases.
The tuberculosis results also covered latent TB, which has no symptoms but can develop into active disease. Haick said the team first promised Bill Gates a manufacturing cost of $1; the patch now costs a little over a dollar. Those figures are specific. The unanswered question is whether the device will keep performing outside trials and laboratory conditions.
Haick’s other diagnostic system, an electronic nose about the size of a hand, analyzes volatile organic compounds in exhaled breath. He said it can identify 17 diseases and is about 88% accurate, but the U.S. Food and Drug Administration gave it Breakthrough Device designation for two conditions: lung cancer and advanced liver fibrosis unrelated to alcohol consumption. Hospitals around the world are using it experimentally for those conditions.
A promising signal isn’t the same as a dependable diagnosis. Prof. Jonathan Sleeman of Heidelberg University said clinical validation needs large patient groups, and researchers must establish whether different diseases can produce the same positive result. “The arguments he presents are compelling, but that has to be shown in the real world, and that's where we're going,” Sleeman said.
Funding also determines which research gets room to continue. Prof. Sylvia Cohen-Kaminsky said an electronic-nose project on pulmonary arterial hypertension distinguished patients from controls and identified people with a genetic mutation associated with increased susceptibility. The work stopped there after researchers failed to secure follow-up funding. Haick said a European Union grant awarded in October 2006 provided 1.73 million euros, allowing him to establish a laboratory with 17 students and researchers rather than two. EU-funded consortia now include laboratories, hospitals and commercial companies; one lung cancer trial has 5,600 participants and a 14.8 million-euro grant. Money builds capacity. It doesn’t replace validation.
AI training, with a rights question attached
In Lebanon, the “One Million Lebanese AI Experts” initiative aims to familiarize up to one million people with generative AI. Supported by the UAE, officials announced it during talks between Sheikh Abdullah bin Zayed Al Nahyan and Prime Minister Nawaf Salam on the sidelines of the United Nations General Assembly. Minister of the Displaced and Minister of State for Technology Affairs Kamal Sehahdeh said the project builds on a Lebanese strategy developed “for nearly a year.”
The planned training consists of three-hour sessions in four modules covering fundamental and practical generative-AI and prompt skills. The initiative is still developing. Digital rights advocacy group SMEX has raised concerns, though the available account doesn’t specify them. A state-backed training program can count its planned reach in millions. The rights questions remain part of the story, not a footnote to the rollout.
Models aren’t evidence of a cure
In Abu Dhabi, Insilico Medicine researchers developed LongevityBench, a benchmark for assessing AI reasoning across biological data related to aging. Published in Cell in September 2026, it covers 17 tasks using clinical, epigenetic, transcriptomic, proteomic and genetic data. The researchers tested systems from OpenAI, Google, Anthropic, xAI, DeepSeek and Moonshot AI, alongside five specialized Longevity-LLMs. The specialized models matched or exceeded most of the frontier systems.
“This is the finding I find most important, and it goes against the prevailing assumption in AI that bigger always wins,” said Alex Zhavoronkov, founder and CEO of Insilico Medicine. The benchmark doesn’t settle how biological aging should be defined or measured, he said. Its measures include chronological age, mortality or survival, and established biomarkers; “These are not perfect proxies for the full biological process of aging,” he said.
Longevity Claw nominated 328 genes in an initial autonomous research campaign, with up to 5.6-fold enrichment against an independently published set of experimentally supported aging targets, Zhavoronkov said. The proposed targets still need wet-lab validation. He also said AI target predictions and aging clocks are tools for generating hypotheses, not proof of efficacy or safety in humans. Whether the technology is a low-cost patch or a system trained on biological data, its claims still have to meet the evidence.