Buzzworthy developments of the past few days.

• Anything you say online can and (probably) will be used to help Google train generative AI. The Big Tech heavyweight said as much when it updated its privacy policy this week. Where once Google said it would only use language-based AI for translating text, it now states its intent to use “publicly available information” to help train its AI models as well as “build products and features like Google Translate, Bard and Cloud AI capabilities.” NewsNation coverage here.
   
• Explainability in medical AI is overrated. That’s one way to paraphrase one opinion of healthcare AI thought leader Marzyeh Ghassemi, PhD. In a recent “Grand Rounds” podcast hosted by NEJM AI, the MIT prof makes the case that explainability can be a “technically squishy” concept as well as a red herring: Sometimes black-box AI produces better outputs. Listen here.
   
• AI is helping researchers safely induce allergen tolerance in people whose severe allergic reactions are triggered by edible things. The work is happening at the Food Allergy Institute of Long Beach, Calif. “Patients enter the process with [potentially deadly] anaphylaxis,” explains Inderpal Randhawa, MD, founder of the center. “Thanks to the AI-powered Tolerance Induction Program methodology, they leave the program with what we call ‘food freedom.’” Journalistic video here.
   
• The U.N. Security Council is circling AI. The august body will convene a meeting July 18 to discuss threats the technology could pose to international peace. AP coverage here.
   
• Only one health app ranks among the ‘most invasive AI apps’ on offer in Apple’s App Store. That would be K Health. The operation compiling the list, Home Security Heroes, independently reviewed 149 apps to see which ones were most aggressive about tracking users’ personal info for marketing purposes. Details here.
   
• Say hello to the ‘first AI-driven video library for the medical industry.’ The fruit of a partnership between the AI video management company AnyClip (New York) and the medical conference organizer Kenes Group (Tel Aviv), the library lets clinicians and researchers find granular content with simple search steps. Announcement.
   
• Health insurers have a new AI-based toolkit to help ease their toils. Wisedocs of Toronto says its AI Medical Summary Platform builds on an earlier version to let users quickly synopsize thousands of pages of medical records for fast decision-making on claims and such. Announcement.
   
• Did you know training a single AI model can suck up more electricity than powering 100-plus U.S. homes for a year? Pymnts.com breaks down the problem, explaining why AI’s thirst for fuel is a growing concern to energy companies and environmentalists alike.