AI Morning Post | May 12th one ️⃣ Google first uses AI to create zero day vulnerabilities Google Threat Intelligence has released a report, discovering for the first time that hackers are using AI to independently develop zero day vulnerabilities and are preparing to exploit them on a large scale. Organized black industries use AI to generate attack code, bypass two factor authentication, and even leave fictional CVSS scores generated by AI in the code - this "non-human" trace has become evidence instead. Google blocked the vulnerability before the attack landed. two ️⃣ Google Android Show starts broadcasting today Google's pre I/O live stream went live at 10am PT today. The official version of Android 17 will be released next month, and Gemini will be integrated into the operating system from the bottom layer, supporting lock screen voice call agents to autonomously execute tasks. More noteworthy is Aluminum OS - Google's Android based PC system, which is highly likely to have hardware partners announcing synchronization on I/O. three ️⃣ South Korean startup RLWRD trains robot AI with worker movements AP reported on the methodology of this South Korean company: instead of writing code, workers are required to repeatedly perform the same operation (folding napkins, screwing screws), and the complete operation behavior is recorded as data and fed to the robot. It sounds simple, but the actual effect is better than traditional reinforcement learning, especially in fine-grained operation scenarios. four ️⃣ Swedish AI Cafe Crash Record Nordic experiment results: AI agent "Mona" used Gemini to manage coffee shop operations and ordered 3000 rubber gloves, 6000 napkins, 4 first aid kits, and either ordered too much or forgot to order bread. In the first month of operation, the revenue was $5700 and the cost exceeded $21000. The human barista remained speechless throughout the entire process. ——The technology is cool, but the implementation is cruel. five ️⃣ Wearable devices+AI for predicting health Oura, Fitbit, and Whoop have started to integrate AI analysis of biochemical indicators, no longer simply recording steps and heart rate, but providing warnings before abnormalities occur. This may be the most practical evolutionary direction for health management products. Summary: AI security has gone from theory to practice, and Google has sounded the alarm; On the other hand, the pitfalls of product landing are still being stepped on. Both lines are accelerating.