Israeli chipmaker Hailo Technologies has released the Hailo-10H, the first discrete AI accelerator designed for generative AI workloads at the edge. The device runs large language models (LLMs), vision-language models (VLMs), and other multi-modal AI directly on-device, eliminating the need for cloud-based inference. The Hailo-10H offers unmatched power efficiency and low latency, achieving first-token generation in under one second and maintaining 10 tokens per second on 2-billion parameter LLMs. It can also generate images with Stable Diffusion 2.1 in under five seconds, demonstrating a significant leap forward for offline generative workloads. The chip is designed around Hailo’s second-generation neural core architecture, providing 40 tera-operations per second (TOPS) of INT4 performance and 20 TOPS of INT8 at a typical power draw of 2.5 W. It is fully compatible with TensorFlow, PyTorch, ONNX, and Keras, and is supported by Hailo’s mature software stack. The device is designed to work in hybrid AI pipelines that blend LLMs or VLMs with traditional convolutional neural networks (CNNs), conserving power and ensuring real-time responsiveness for mission-critical applications like video analytics.
Anthropic unveils ‘auditing agents’ to test for AI misalignment finding prompts that elicit “concerning” behaviors
Anthropic researchers developed auditing agents that achieved “impressive performance at auditing tasks, while also shedding light on their limitations.” The researchers stated that these agents, created during the pre-deployment testing of Claude Opus 4, enhanced alignment validation tests and enabled researchers to conduct multiple parallel audits at scale. The three agents they explored were: Tool-using investigator agent for open-ended investigation of models using chat, data analysis and interpretability tools; Evaluation agent that builds behavioral evaluations and can discriminate between models that have implanted behaviors and those that do not; Breadth-first red-teaming agent, which was developed specifically for the Claude 4 alignment assessment, so that it can discover implanted test behaviors. According to the researchers, the investigator agent successfully found the root cause of the issues 10-13% of the time. Still, this performance improved to 42% “when using a super-agent approach that aggregates findings across multiple investigations.” The next step in most alignment audits is typically a behavioral evaluation, which determines whether the model consistently misbehaves in various settings. To test its evaluation agent, the researchers designed 10 models “given system prompt instructions to exhibit a specific behavior, such as excessive deference to the user.” They ran the agent five times per model and saw that the agent correctly finds and flags at least one quirk of the model. However, the agent sometimes failed to identify specific quirks consistently. It had trouble evaluating subtle quirks, such as self-promotion and research-sandbagging, as well as quirks that are difficult to elicit, like the Hardcode Test Cases quirk. The last test and agent concern behavioral red-teaming to find the prompts that elicit “concerning” behaviors. The breadth-first red-teaming agent converses with the target model (in Anthropic’s case, it was with Claude Opus 4), and this chat is then rated for alignment-relevant properties. The agent identified seven of the ten system quirks, but it also struggled with the same issues as the evaluator agent.
Custody platforms are emerging as the core infrastructure for governing token operations through programmable rules, multi-layer approvals, role-based access, and integration-ready APIs for institutions navigating hybrid environments
Custody is the core infrastructure layer that makes institutional participation in digital assets not just possible but profitable. Modern custody platforms serve as critical mission-control layers, orchestrating transaction workflows, enforcing policy, supporting compliance and enabling the tokenization of real-world assets (RWAs) at scale. Ripple Custody is built for this new paradigm, enabling institutions, fintechs, and crypto businesses to govern token operations through programmable rules, multi-layer approvals, role-based access, and integration-ready APIs. Security is table stakes with Ripple’s system being ISO 27001-certified, SOC 2 audited, and supporting the most rigorous key management architectures on the market. However, its flexibility sets it apart, as it supports a range of deployment options, including on-prem, SaaS, or a hybrid of the two. This allows financial institutions to align custody models with internal policy, client demands, and jurisdictional requirements. The market for real-world asset tokenization is accelerating fast, with forecasts projecting the tokenized RWA market will grow from $0.6 trillion in 2025 to nearly $19 trillion by 2033. Ripple Custody is built with this reality in mind, enabling institutions to define multi-party approval flows, enforce transaction policies down to asset, role, and time-based rules, and lock in governance frameworks that reduce the risk of social engineering, collusion, and unapproved access. Bridging TradFi and Web3, Ripple Custody bridges both crypto-native and TradFi-oriented platforms. With over a decade of experience operating across crypto markets and institutional finance, Ripple brings a dual advantage: a deep understanding of blockchain infrastructure and a real fluency in bank-grade systems. Custody enables what comes next, as institutions move from tokenization theory to practice. It’s not about picking a vault; it’s about selecting a partner that can evolve with your business as strategies mature. Ripple Custody offers institutions a platform, not a black box, and strategic enablement. Custody systems must now support: Tokenized securities and bonds, with customizable workflows and strict governance; Money Market Funds and private credit, which demand real-time settlement, role-based access and policy-based controls; Stablecoins, with over $215 billion in circulation as of mid-2025, and growing relevance in cross-border treasury flows.
Research shows spin polarization property in gold nanoclusters can be easily synthesized in relatively large quantities to support and scale a variety of quantum applications
A team of researchers from Penn State and Colorado State has demonstrated how a gold cluster can mimic gaseous, trapped atoms, allowing scientists to take advantage of these spin properties in a system that can be easily scaled up. The researchers show that gold nanoclusters have the same key spin properties as the current state-of-the-art methods for quantum information systems. They can also manipulate an important property called spin polarization in these clusters, which is usually fixed in a material. These clusters can be easily synthesized in relatively large quantities, making this work a promising proof-of-concept that gold clusters could be used to support a variety of quantum applications. An electron’s spin not only influences important chemical reactions but also quantum applications like computation and sensing. The direction an electron spins and its alignment with respect to other electrons in the system can directly impact the accuracy and longevity of quantum information systems. Gold clusters can mimic all the best properties of the trapped gaseous ions with the benefit of scalability. Scientists have heavily studied gold nanostructures for their potential use in optical technology, sensing, therapeutics, and to speed up chemical reactions, but less is known about their magnetic and spin-dependent properties. In the current studies, the researchers specifically explored monolayer-protected clusters, which have a core of gold and are surrounded by other molecules called ligands. The researchers determined the spin polarization of the gold clusters using a similar method used with traditional atoms. The research team plans to explore how different structures within the ligands impact spin polarization and how they could be manipulated to fine tune spin properties. This presents a new frontier in quantum information science, as chemists can use their synthesis skills to design materials with tunable results.
Meta is developing a wristband that lets people control a computer using hand gestures that includes moving a cursor, opening apps, and sending messages by writing in the air as if using a pencil
Meta researchers are developing a wristband that lets people control a computer using hand gestures. This includes moving a cursor, opening apps, and sending messages by writing in the air as if using a pencil. Meta’s wristband employs a technique called surface electromyography (sEMG), which detects electrical signals generated by muscle activity to interpret user movements, as explained in a research paper published in the journal Nature. These signals can sense a person’s intended actions, even before they do them. The goal of this wristband is to provide less invasive tools to interact with computers for people with motor disabilities. The company is collaborating with Carnegie Mellon to test the wristband with people who have spinal cord injuries, enabling them to use computers even if they are unable to fully use their arms or hands. Douglas Weber, a professor in the Department of Mechanical Engineering and the Neuroscience Institute, noted that even people with complete hand paralysis still exhibit some muscle activity. As a result, the device is able to interpret their intended actions. The wristband is a simpler alternative to other invasive projects such as Elon Musk’s Neuralink. Meta’s wristband could be used immediately without surgical intervention, and its use of sEMG signals means it operates at a higher frequency than EEG.
Pulsate’s mobile-first engagement platform taps customer intent signals to enable community banks recognize and act on customer micro-engagements through delivering highly personalized, differentiated and contextual offers directly within existing banking platforms
Sarah Martin is CEO at Pulsate, a mobile-first, personalized consumer engagement platform tapped largely by community banks and credit unions. Pulsate’s data, for instance, capitalizes on banking customer intent signals to recommend next-best products that convert interest into action. Martin says to compete in a widening financial services market, institutions of any size must believe they too can attract and retain customers, drive deposit growth and loan balances, as well as react to dynamic market fluctuations just like their neobank competition, money centers and others. Regional banks, community banks, credit unions and banks of all sizes can expand their approach and smartly use the resources they have by thinking as their customers do. Yes, data and technology are key, she says, but institutions should strategize how they can best leverage in-house capabilities, often with partnerships. She says the design of her platform targets a specific goal, however, one she believes too many financial institutions take for granted: recognizing and acting on customer micro-engagements. Finding value not just in product, but in service, in overall experience, in empathy, in message, in brand, in differentiation, in delivering highly personalized, contextual offers and communications directly within a bank or credit union’s existing mobile and digital banking platforms is what Martin has found can effectively transform a mediocre or even satisfactory digital or mobile banking channel into a comprehensive profit center. Every encounter is an opportunity. Feedback is data and data informs what can happen next.
Credibur’s platform orchestrates the full lifecycle of institutional funding for alt lenders from structuring through reporting and contract management, to capital calls and administration of special purpose vehicles (SPVs) through a modular API-and-AI-first infrastructure
Credibur has secured $2.2 million (€1.85 million) in pre-seed funding to launch its credit infrastructure platform. With Credibur, Nicolas Kipp, Founder and CEO of Credibur, and his team are developing infrastructure that solves a central problem in the structured credit portfolio business between non-bank lenders and institutional capital providers: it automates complex, Excel-based workflows and provides all critical decision-making data in real-time. The new platform for debt facility management goes far beyond traditional reporting tools, orchestrating the full lifecycle of institutional funding: from structuring through reporting and contract management, to capital calls and the administration of special purpose vehicles (SPVs). The solution targets alternative lenders such as BNPL providers, factoring and leasing companies, as well as institutional investors, including asset managers, debt funds, and family offices. Credibur’s modular API-and-AI-first infrastructure enables a more informed risk assessment and improved decision-making in the credit business. Time-consuming, error-prone Excel lists managing millions of euros become a thing of the past, as data is delivered directly from systems via interfaces. With Credibur, Kipp is now solving the fundamental problem: manual debt facility management is slowing growth across the entire private credit sector. His infrastructure can finally digitalise this €430 billion industry in Europe.
LinePoint Partners platform enables ultra-high-net-worth (UHNW) breakaway financial advisors and single family office (SFO) executives to independently operate and grow their practices without designing from scratch in-house
One new entrant to the family office industry, LinePoint Partners, aims to help advisors leaving wirehouses and private banks build up an array of services needed to work with wealthy clients. Andrew Sternlight, LinePoint’s president and chief investment officer, said what sets family offices apart from other wealth managers is the priority family offices place on preserving wealth for future generations, charitable causes or other recipients. “That’s where we think, by extending that sort of family office infrastructure to advisors or to one or two executives of a family office without building out their teams entirely,” Sternlight said. “That’s where we can provide a bit of a solution that has the benefits with that multigenerational lens, but not the cost of designing from scratch in-house.” Andrew Lom, U.S. head of financial services and global head of private wealth at the law firm Norton Rose Fulbright, agreed that the true differentiator for family offices is the emphasis they place on making sure clients can bequeath their wealth exactly how they want. Ron LaVelle, a principal in the private wealth practice of the accounting and consulting firm Baker Tilly, said clients have the ultimate say on what their family office does. With the list of possible service offerings always growing, and the number of multifamily offices serving several wealthy families also rising, it’s becoming ever more rare for firms to be able to do it all. In fact, LaVelle said, he tends to be wary of firms that claim to be everything for everyone.
FedNow enters year three with increasing users, volumes, and competition, reporting a whopping 1,200% year-over-year increase in transaction volume, growing from 97,424 settled payments in the first quarter of 2024 to 1,310,017 in the quarter ending March 31, 2025
FedNow, which turned two years old on July 20, reported a whopping 1,200% year-over-year increase in transaction volume, growing from 97,424 settled payments in the first quarter of 2024 to 1,310,017 in the quarter ending March 31, 2025. At its one year anniversary last July, the Fed had enrolled more than 850 financial institutions into FedNow and had at least 1,000 more in the pipeline. But as of July 7, more than 1,400 financial institutions—including large and small banks and credit unions—were participating in the FedNow Service. Community banks and credit unions make up more than 95% of the platform’s total participants. That all sounds impressive until you look at data recently reported by competitors RTP and Zelle. The RTP network, which is operated by The Clearing House and owned by multiple large banks, makes FedNow’s transaction volume look tiny by comparison. RTP handles more than 1 million daily transactions and, in fact, set a new single-day record for payments volume at nearly 1.6 million transactions on Jan. 31. The RTP network currently reaches 70% of demand deposit accounts in the United States, meaning that millions of consumers and businesses are benefiting from instant payments through the 850 financial institutions connected to the network. Tim Scholten, founder and president of the credit union and community bank consultancy Visible Progress, told Tyfone that FedNow adoption is growing but slowly.
Scrunch AI’s platform enables brands to serve compressed, structured, and machine-readable content specifically for AI crawlers and agents, thereby improving discoverability and placement without requiring a redesign of existing websites
Scrunch AI, the company helping brands understand and improve how they appear in AI search, announced a $15 million Series A funding round led by Decibel, with participation from Mayfield, Homebrew, and other strategic investors. The company is powering the shift from traditional SEO to AI search optimization. In addition to its enterprise-grade solutions, Scrunch has also launched a dedicated offering for agencies, designed to help them guide their clients through the transition to AI search. This program provides tailored access, insights, and tools specifically built for managing multiple brands and implementing at scale. The newly raised capital will accelerate the development and deployment of the Scrunch Agent Experience Platform (AXP), a new infrastructure layer that helps brands serve optimized content directly to AI crawlers and agents. Traditional websites designed for human browsers are often inefficient and ineffective for AI. AXP enables brands to maintain their existing websites for users while delivering compressed, structured, and machine-readable content specifically for AI engines, thereby improving discoverability and placement without requiring a redesign. AXP is currently in pilot with enterprise customers and will be rolled out more broadly in the months ahead.