Welcome back to Daily Zaps, your regularly-scheduled dose of AI news ⚡

Here’s what we got for ya today:

Let’s get right into it!

Sign Up | Advertise | Tools

New York City will temporarily ban elementary and middle school students from using generative artificial intelligence tools during the upcoming school year. The one-year ban applies to students through eighth grade and is intended to give education officials time to study how AI affects learning. Meanwhile, high school students will receive AI literacy lessons twice a year, and some will be allowed to use approved AI tools under teacher supervision. Companion chatbots will be banned for students of all ages. Teachers, however, will still be permitted to use AI for lesson planning and other work-related tasks.

The city is also recommending limits on students’ screen time, including 30 minutes per day for grades three through five and 45 minutes for grades six through eight. Officials plan to review technology currently used in schools and remove tools that do not contribute significantly to learning. Supporters believe these policies could prevent students from becoming too dependent on AI or using it to cheat, while critics argue that one year may not be enough to understand AI’s long-term effects. Overall, the city is trying to balance the educational benefits of AI with concerns about student safety, academic honesty, and face-to-face learning.

OpenAI has classified its new AI model, Astra, at the “Critical” cybersecurity risk level, the highest level in the company’s safety framework. Astra can independently discover previously unknown security weaknesses, known as zero-day vulnerabilities, and develop methods to exploit them. During testing, Astra earned a perfect score on one cybersecurity benchmark and discovered two new zero-day vulnerabilities. It also demonstrated the ability to combine multiple security flaws into complete attacks against protected computer systems with little or no human assistance.

Because of Astra’s advanced capabilities, OpenAI has introduced stronger safety measures to prevent the model from being misused. The company temporarily paused parts of Astra’s training and delayed its release while improving security, monitoring, and the model’s ability to reject harmful requests. Astra’s most powerful cybersecurity features will not be released to the general public at first and will instead be limited to selected testers focused on defensive security. The development of Astra shows that AI could help cybersecurity experts find threats faster, but it also raises concerns about how quickly attackers could exploit vulnerabilities if similar technology is misused.

Moyai evaluates agent runs and tell you why they broke.

Stop guessing if your agents are failing and start monitoring them for failures. We surface agent failures by looking for behavioural anomalies your your agent traces and classify failures with RCA and remediation steps. Works on your existing observability stack, no new SDK required.

Talk to Founder

Google’s Gemini 3.8 Flash introduces major improvements in coding and software engineering compared with Gemini 3.7 Flash. The model is designed to handle long, complex coding projects with less human assistance, allowing it to solve problems from beginning to end. It performs better than many larger and more expensive AI models on tests involving long-term software engineering. Gemini 3.8 Flash also takes additional reasoning steps and repeatedly uses tools to check and improve its work. These abilities allow it to build complex programs from simple instructions, including interactive games, mapping applications, and 3D visualizations.

Gemini 3.8 Flash Cyber also improves coding by focusing on finding and fixing security problems in software. It can search large codebases across 20 programming languages and successfully identify more than 70% of vulnerabilities in Google's internal testing. The model is also capable of automatically creating patches to repair security flaws. According to Google, it produced 2.6 times more correct security patches for Chrome than larger commercial AI models. These improvements show that Gemini 3.8 can not only generate complex code but also analyze, repair, and secure existing software more effectively.

Tesla plans to expand its Robotaxi service to about a dozen states by the end of the year as part of the company’s move toward fully autonomous transportation. The company is developing the Cybercab, an electric vehicle designed to operate without a human driver. The Cybercab is expected to have all-wheel drive, 219 horsepower, and a 48 kWh battery with an estimated range of about 418 miles. Tesla currently operates Robotaxis using Model Y vehicles in limited areas and plans to expand to additional cities. CEO Elon Musk has said that safety testing and careful validation will be important before the service expands further.

Riding in a Tesla Robotaxi will be similar to using services such as Uber or Lyft, except there will be no driver behind the wheel. Passengers will use Tesla’s mobile app to request a ride, wait at a designated pickup location, and receive a notification when the vehicle arrives. Once inside, passengers will press a “Start Ride” button and can adjust settings such as the climate and seat position. Tesla plans to offer simple introductory pricing, although prices may change in the future. Overall, the Cybercab represents Tesla’s effort to make fully autonomous vehicles a larger part of everyday transportation.