NerdNews

Palo Alto Networks warned customers that a critical-severity PAN-OS firewall zero-day vulnerability, tracked as CVE-2026-0300, has been exploited by suspected state-sponsored hackers for nearly a month. The vulnerability allows unauthenticated attackers to execute arbitrary code with root privileges on Internet-exposed PA-Series and VM-Series firewalls.

Meta is working on an AI agent called Hatch, inspired by OpenClaw, to enable agentic shopping on Instagram and compete with TikTok Shop. The agent will work within Meta's apps and with outside services, aiming to make AI more accessible and deliver 'superintelligence' capabilities.

OpenAI has launched new voice intelligence features in its API, including GPT-Realtime-2 for conversational simulations, GPT-Realtime-Translate for real-time translations, and GPT-Realtime-Whisper for live speech-to-text capabilities. These features can be used in various applications such as customer service, education, and media.

Spotify is introducing a new feature that allows users to create and import AI-generated podcasts into the app using programming tools like OpenAI's Codex or Anthropic's Claude Code. These podcasts will appear in the user's library for easy access, but won't be available to other users.

Anthropic's developer conference featured the announcement of a deal with SpaceX, new Managed Agents features, and insights from the company's platform team. The deal with SpaceX allocates all capacity in the Colossus supercluster to Claude, increasing rate limits and removing peak-hour limits. Managed Agents features include multi-agent orchestration, dreaming, and outcomes. The platform team discussed the end of generic harnesses, infrastructure challenges, and the need for agent babysitters.

OpenAI has launched a Chrome extension for its Codex platform, allowing users to test web apps, collect context from open tabs, and use Chrome DevTools in parallel. This expansion aims to make Codex more appealing to casual users and additional professions beyond developers.

OpenAI introduced a WebSocket-based execution mode for its responses API, reducing latency by up to 40% and improving throughput in high-concurrency scenarios. This change replaces traditional HTTP request-response patterns with a persistent, bidirectional connection, enabling continuous data exchange and faster tool execution.

Adam Wolff shares lessons from building Claude Code, an AI-powered coding tool, highlighting the importance of experimentation, feedback loops, and optimizing delivery cadence in AI-driven software development

Google has announced GKE Agent Sandbox for secure agent code execution and GKE Hypercluster for managing up to a million accelerator chips. These updates position Kubernetes as an AI agent, with GKE now powering AI workloads for top customers.

A new study suggests that using AI chatbots for as little as 10 minutes can have a negative impact on people's ability to think and problem-solve. Researchers found that participants who used AI assistants were more likely to give up on problems or provide incorrect answers when the AI help was suddenly taken away.

A former OpenAI employee testified that the company's focus on product development compromised its commitment to AI safety, sparking concerns about the organization's priorities and governance structure. The lawsuit, brought by Elon Musk, alleges that OpenAI's transformation from a research organization to a for-profit company has led to a shift away from its founding mission of ensuring that humanity benefits from artificial general intelligence.

Anthropic's Mythos model has significantly improved Firefox's cybersecurity by discovering thousands of high-severity bugs, including some that had been dormant for over a decade. The model's capabilities have shifted the advantage to defense, but its impact on the broader balance of power in cybersecurity is still unclear.

Google's AI health coach, part of Google Health Premium, offers personalized fitness and wellness guidance for $9.99/month, leveraging Gemini AI and available on May 19.

Mozilla has fixed 423 Firefox security bugs in April, with the help of AI model Mythos. The model found 271 of these bugs, but it's unclear if the model or better middleware is responsible for the improvement. Mozilla has released a dozen examples of the bugs found, including a 20-year-old heap use-after-free bug.

AMD's new MI350P GPU offers 144 GB of HBM3e and up to 4.6 petaFLOPS of FP4 compute, competing with Nvidia's H200 NVL and RTX Pro 6000

Google has launched the Fitbit Air, a screenless wearable device that tracks fitness and health metrics. The device is part of Google's new health initiative, which includes a consolidated health app and an AI-powered health coach. The Fitbit Air is priced at $99 and is available for purchase.

Mira Murati, former OpenAI CTO, played a significant role in Sam Altman's ouster. Her deposition revealed that she had concerns about Altman's management style and had shared these concerns with the board. However, she later supported Altman's reinstatement, causing confusion about her stance.

Anthropic has signed a deal with SpaceX to use computing resources from xAI's data center, including the Colossus 1 supercomputer. This partnership will provide Anthropic with over 300 megawatts of new capacity and approximately 220,000 Nvidia GPUs, directly improving capacity for Claude Pro and Claude Max subscribers.

Moonshot AI, a Beijing-based AI lab, has raised $2 billion at a valuation of $20 billion. The company's open-weight large language models, including Kimi K2.5 and Kimi K2.6, have gained popularity and compete with models from OpenAI, Google, and Anthropic.

HPE has released its first collaboration between Juniper and Aruba, featuring self-driving Wi-Fi access points that use AI to autonomously manage networks. The HPE Networking 723H access points can be managed with either Aruba Central or the Mist platform and include features like dynamic capacity optimization and rogue DHCP server detection.