NerdNews

A new benchmark developed by Salesforce academics shows LLM-based AI agents perform below par on standard CRM tests, achieving a 58% success rate on single-step tasks and 35% on multi-step tasks, highlighting concerns over their ability to handle confidential information.

Microsoft is at an impasse with OpenAI due to a dispute over access to Windsurf's intellectual property. OpenAI doesn't want Microsoft to have access to Windsurf's IP, while Microsoft currently has access to all of OpenAI's IP. This disagreement is causing tensions between the two companies.

Alibaba has launched an upgraded variant of its Qwen3 AI model, compatible with Apple's platforms, signaling the debut of Apple Intelligence in China. The model supports 119 languages and features hybrid architecture and better performance. Apple Intelligence is expected to launch in China with the release of iOS 18.6.

OpenAI and Microsoft's relationship is reportedly deteriorating due to concerns over anticompetitive behavior, intellectual property, and computing resources. OpenAI is trying to loosen Microsoft's grip on its assets, but needs the tech giant's approval for its for-profit conversion. The two companies are in a standoff over OpenAI's $3 billion acquisition of Windsurf, with OpenAI not wanting Microsoft to gain Windsurf's intellectual property.

Over 22,000 workers have been laid off in the tech industry in 2025, with 16,084 cuts in February alone. Companies like Microsoft, Amazon, and Google have made significant layoffs, affecting various roles and departments.

The AirPods Pro 3 may be delayed until 2026 due to Apple's struggles with AI, particularly with its Siri upgrade. The new AirPods are expected to feature major upgrades, including a health sensor and potentially cameras. The delay is likely tied to Apple's efforts to perfect its AI technology before releasing new products.

The article discusses the importance of automated guardrails in vibe coding, which is developing software without automated checks and quality gates. It highlights the benefits of using tools like pre-commit hooks, CI pipelines, and automated release management to ensure code quality, security, and maintainability. The author provides a real-world example of a command-line tool that creates acronyms from phrases, demonstrating how automated guardrails can be implemented in a development workflow.

US Customs and Border Protection has the authority to search travelers' devices. To protect yourself, consider using a travel phone with a sanitized version of your digital life, or modify your primary phone by removing old photos and messages. Keep your phone's operating system up to date and use a PIN or alphanumeric code instead of biometrics. Be prepared to refuse a device search, but be aware of the potential consequences.

Apple has dramatically upgraded Personas in visionOS 26, making them look more realistic. The new Personas have improved facial geometry, hair, and expressions, and are no longer blurry or stiff. They can be used for video calls and other applications, and Apple is expected to release visionOS 26 in the fall.

The article discusses the importance of strategic thinking for staff+ engineers, providing a framework for cultivating this skill. It highlights the need to balance risk and innovation, and offers real-world examples of strategic thinking in action. The author also emphasizes the role of leadership in empowering staff+ engineers and provides tips for influencing organizational culture.

Tesla's Full Self-Driving mode has failed tests by blowing past a stopped school bus and hitting kid-sized dummies. The tests, conducted by The Dawn Project, showed the software repeating the same mistake eight times, raising concerns about the readiness of fully autonomous cars.

The PCIe 7.0 specification has been released, delivering a theoretical maximum bandwidth speed of 512GB per second in both directions, across a x16 connection. It's targeted at data-driven applications like cloud and quantum computing datacenters, and will be backward compatible with previous PCI Express versions.

Google's Threat Intelligence Group warns of Scattered Spider's ransomware attacks on US insurance companies, following similar attacks on retailers. The group uses social engineering schemes, including fake help-desk calls, to gain access to systems. Multiple insurers' networks are still down after the attacks.

Alta, a startup founded by Jenny Wang, has raised $11 million in seed funding to develop an AI-powered personal styling agent. The platform allows users to upload their closet and receive outfit recommendations based on their budget, lifestyle, and calendar. Alta's AI technology is trained by fashion experts, including Michelle Obama's stylist Meredith Koop. The company plans to use the funding to grow its team and expand its partnerships with retailers and fashion brands.

The US Department of Defense has contracted OpenAI for a $200 million pilot program to develop 'frontier AI' capabilities, with the goal of addressing critical national security challenges. The project will focus on transforming administrative options and supporting proactive cyber defense, although the exact details of the project are unclear.

European law enforcement agencies have shut down Archetyp, one of the largest dark web drug marketplaces, and arrested its suspected administrator. The marketplace had over 600,000 users and a total transaction volume of at least €250 million. The operation, led by Germany's Federal Criminal Police Office, involved multiple countries and resulted in the seizure of numerous assets, including smartphones, computers, and narcotics.

Microsoft has introduced a new Recall Export feature for Windows 11 users in the European Economic Area, allowing them to share Recall snapshots with third-party apps and websites. The feature is currently rolling out to Windows Insiders and will require users to store an export code safely, as losing it will result in the deletion of all previously saved snapshots.

Meta has added a warning to its AI app, advising users to avoid sharing personal or sensitive information. The move comes after users were found to be sharing intimate details publicly, despite the app's intention to keep chats private. The warning prompts users to be cautious when sharing interactions, reminding them that their prompts may be visible to everyone and suggested by Meta on other apps.

Archetyp Market, a major dark web marketplace, has been shut down by a Europe-wide law enforcement operation. The platform, which had been operating for over five years, facilitated over €250m in transactions and offered more than 17,000 listings, including synthetic opioids. The operation, dubbed Operation Deep Sentinel, involved authorities from six countries and resulted in the arrest of the platform's administrator and several other key players.

A series of Freedom of Information requests shows that students in British universities are increasingly getting busted for using AI to cheat, with 7,000 cases reported between 2023 and 2024. AI companies are offering discounts and promotions to students, raising concerns about the role of AI in academic dishonesty.