NerdNews

AWS launches OpenSearch Serverless, a fully managed search and vector database designed for agentic workloads, allowing for instant scalability and cost savings. This shift reflects the growing need for infrastructure to adapt to machine-generated traffic, which is expected to exceed human traffic by 2027.

Anthropic has raised $65 billion in funding at a $965 billion post-money valuation, marking its last private fundraising before debuting on the public markets. The company plans to use the new funds to advance its safety and interpretability research and expand compute to meet growing demand for its AI model, Claude.

General Compute, a new inference neocloud, has raised $15 million in seed funding to address the growing demand for AI computing power. The company is partnering with SambaNova, an Intel-backed chipmaker, to use their specialized SN50 chips, which claim to outperform GPUs and other specialized chips. General Compute aims to provide faster and more cost-effective AI processing power, with plans to deploy the new chips in existing data center facilities.

Apple is working to distill Google's multi-trillion parameter Gemini AI to run on iPhone, but it will likely rely on cloud components from Google and Nvidia, potentially compromising Apple's privacy-focused approach to local AI processing.

Anthropic has confirmed that it plans to bring Mythos-class models to the general public after delaying the rollout due to security risks. The Mythos model shows major improvements in code reasoning and autonomy, far above Claude's current flagship model, Opus 4.8.

Adaptive hedged requests can reduce p99 latency by 74% in distributed systems. This approach learns the latency distribution from live traffic and fires hedges at the right point, preventing load amplification during outages. It's suitable for load-balanced, multi-instance deployments and can be applied to LLM inference workloads.

This article explains how to run a local AI chatbot on an iPhone, highlighting the benefits of cost savings, privacy, and offline usage. It recommends two apps, Locally AI and Private LLM, for easy installation and use of open-source LLMs. The article also discusses the tradeoffs between model complexity, storage space, and performance.

Asana has acquired StackAI, a no-code agent-builder, for $75 million to enhance its AI-native workplace platform. StackAI's founders will join Asana, and the acquisition is part of Asana's broader AI pivot to build an 'operating system for human-agent teams'.

Sesame, a conversational AI startup founded by Oculus founders, has launched its iOS app, offering a new type of chatbot experience with four distinct AI agents, fast search and retrieval systems, and technology that allows it to run multiple parallel searches while speaking.

Director Valerie Veatch discusses her documentary critiquing AI hype narratives, highlighting the technology's harmful effects on labor and the environment, and promoting a culture of technological refusal.

Apple is preparing to reintroduce the new Siri at WWDC 2026, with a redesigned interface that puts the Gemini-powered AI agent front and center. The new Siri will live inside the iPhone's dynamic island and allow users to launch apps, start text messages, and search through notes. Apple is also considering giving users the option to access other AI services through this new interface.

Waymo has introduced its newest robotaxi, the Ojai, a Chinese-made, all-electric minivan designed to lower costs and handle high rider demand. The vehicle is equipped with Waymo's sixth-generation system, including 13 cameras, four lidar sensors, and six radar units. The Ojai is currently available to select riders in Los Angeles, Phoenix, and San Francisco, with plans to expand access to more riders and cities.

Vertu has launched the Alphafold, a foldable smartphone powered by an AI agent that connects with enterprise software and coordinates workflows. The device starts at $6,880 and features a 8.05-inch foldable display, Qualcomm's Snapdragon 8 Gen 4 processor, and a triple rear camera setup. The AI agent, called Hermes Agent, can connect to enterprise systems like ERP and CRM, and coordinate tasks such as approvals, scheduling, and sales tracking.

Anthropic has launched Claude Opus 4.8, the latest version of its AI model, which specializes in catching its own mistakes and pointing them out to users. The model boasts industry-leading scores on tasks like agentic coding and computer use. Additionally, Anthropic has teased the upcoming launch of 'Mythos-class models' with capabilities allegedly on par with those of Mythos, a mysterious model that has been delayed due to its unprecedented power and cybersecurity risks.

Microsoft has launched a revamped version of Microsoft 365 Copilot, offering a cleaner design that loads twice as fast. The update includes a feature called 'progressive disclosure' that presents tools and controls based on the user's prompt, and allows for text formatting directly inside Copilot's prompt box.

GCHQ's director, Anne Keast-Butler, urges UK businesses to prioritize cyber security as AI rapidly changes the threat landscape. She warns of a narrowing window to stay ahead of technology and emphasizes the need for immediate action to protect against AI-powered attacks.

Glean, an enterprise AI search company, has reached $300 million in annual recurring revenue, a three-fold increase from the $100 million milestone it reached 15 months ago. The company's AI tools have a deep understanding of customers' business needs, helping enterprises cut AI computing costs, which has become a major selling point.

The Shanghai Futures Exchange is designing a derivatives market for AI tokens, while CME Group and Intercontinental Exchange are working on launching futures contracts for renting GPUs, allowing businesses to hedge against compute costs.

The Verizon 2026 Data Breach Investigations Report highlights that attackers are still winning through access, with credentials deciding how far many attacks can go. Exploited vulnerabilities are now the most common initial access vector for breaches, and credential abuse appears in 39% of breaches. The report emphasizes the importance of managing credentials and non-human access to prevent attacks.

Researchers found that Large Language Models (LLMs) tend to believe false statements even when explicitly warned that they are false. This 'negation neglect' effect occurs when LLMs are fine-tuned on training data that includes false claims, even if those claims are clearly labeled as false. The effect persists even when the negations are repeated numerous times or presented as fictitious or from an unreliable source.