NerdNews

Open source developers are struggling with aggressive AI crawlers that are overwhelming their sites, forcing them to block entire countries. The crawlers, which are used by AI companies to collect data, are ignoring standard blocking measures and causing significant strain on infrastructure. Developers are fighting back with custom-built proof-of-work challenge systems and other defensive tools.

The source code for AlexNet, a convolutional neural network that revolutionized the AI field in 2012, has been released by Google and the Computer History Museum. The code, which is now available on GitHub, represents a key moment in computing history and provides insight into the development of deep learning techniques.

OpenAI's ChatGPT now leverages GPT-4o to natively create and modify images and photos, offering more accurate and detailed results. The feature is available for Pro plan subscribers and will roll out to other users soon.

Google has unveiled Gemini 2.5, a next-gen family of AI reasoning models that can 'think' before answering questions. The model outperforms previous models in several benchmarks, including code editing and multimodal tests. Gemini 2.5 Pro Experimental will be available on Google AI Studio and the Gemini app for subscribers.

Google has rolled out Gemini's Project Astra-powered camera and screen share capabilities, allowing users to share their screen or use the live camera to get Gemini's input on objects around them. This feature is available for Gemini Advanced and Google One AI Premium plan users, giving Google an edge over Apple's limited Visual Intelligence feature.

The increasing power consumption of AI models has led researchers to explore ways to reduce energy usage. Techniques such as pruning, quantization, and optimizing data center operations are being developed to make AI more energy-efficient. However, the lack of transparency from companies like Google and Open AI makes it difficult to accurately estimate power consumption. Researchers are working on benchmarking AI models and developing new technologies like photonic chips to improve energy efficiency.

This article provides a step-by-step guide on how to check if your online accounts, including Gmail, Facebook, Apple, and more, have been hacked. It covers how to view recent activity, log out of suspicious sessions, and enable advanced security features to protect your accounts.

Microsoft has released the first beta of its official Azure SDK for Rust, enabling Rust developers to interact with Azure services. The SDK includes libraries for Identity, Key Vault, Event Hubs, and Cosmos DB. This move signifies Microsoft's recognition of the growing importance and adoption of the Rust programming language.

OpenSilver 3.2, a remake of Microsoft's retired Silverlight, extends the platform to mobile devices using .NET MAUI Hybrid approach, allowing developers to reuse legacy Silverlight or XAML applications on multiple platforms, including iOS, Android, Windows, macOS, web browsers, and Linux.

Otter's new AI Meeting Agent can answer questions and complete tasks based on company meeting data. It can be used on Zoom calls and will be supported on Microsoft Teams and Google Meet soon. Otter also launches Sales Agent and Sales Development Representative Agent.

Tesla's Supercharger network is expanding to multiple EV brands, including GM, Ford, Rivian, and more, with some brands offering adapters for existing customers and others integrating the North American Charging Standard (NACS) into their new vehicles.

Character AI is rolling out parental supervision tools to improve teen safety, including weekly email summaries of activity and time spent on the app. The move comes after lawsuits and criticism for allegedly failing to protect underage users from harm.

Four critical bugs, dubbed 'IngressNightmare', have been discovered in the Ingress NGINX Controller, affecting 40% of cloud systems. The vulnerabilities allow for remote code execution and could enable attackers to take control of targeted clusters. Kubernetes admins are urged to upgrade to patched versions and restrict access to the admission controller.

Researchers found that generative AI browser extensions can collect and share sensitive user data, such as webpage content and personal information, with third-party trackers and servers, potentially violating privacy commitments and regulations like HIPAA and FERPA.

Otter.ai introduces Meeting Agent, an AI tool that can schedule follow-up calls, draft emails, and answer questions based on meeting data. It's part of a suite of AI helpers designed to assist users, with plans to release more 'vertical' agents for marketing, recruiting, and other job functions.

23andMe's genetic data, considered a 'gold mine' for AI companies, is up for sale after the company's bankruptcy. The data could be valuable for training AI models, but its sale raises concerns about privacy and ethics. Companies like Google and OpenAI may be interested in acquiring the data, but face reputational risks due to the sensitive nature of genetic information.

Senior Trump administration officials used Signal to discuss plans to attack Houthi rebels in Yemen, but accidentally added a journalist to the group chat, potentially compromising national security and the lives of military personnel.

Alibaba's Joe Tsai warns that the rush to build AI infrastructure may exceed actual market demand, with companies like Microsoft and Meta investing heavily in datacenters, potentially leading to an unsustainable bubble.

OpenAI has announced that ChatGPT users can now generate images directly within the platform, eliminating the need to switch to Dall-E. This feature is rolling out to all users, including free accounts, and will be the default image generation tool. The platform can generate high-quality images based on prompts, conversations, and uploaded files, and boasts significant improvements in text rendering and contextual understanding.

A threat actor known as EncryptHub has been linked to Windows zero-day attacks exploiting a Microsoft Management Console vulnerability. The vulnerability, tracked as CVE-2025-26633, allows attackers to evade Windows file reputation protections and execute code. EncryptHub has used this exploit to execute malicious code and exfiltrate data from compromised systems, deploying multiple malicious payloads including the EncryptHub stealer and DarkWisp backdoor.