Pulse

Mozilla has reported that Anthropic's Mythos AI model identified 271 security vulnerabilities in Firefox 150. This marks a significant advancement in AI-assisted vulnerability detection, with Mythos producing "almost no false positives" and uncovering flaws that had remained hidden for years. While the AI found numerous bugs, only three were credited as standalone CVEs, indicating a need for continued human oversight in assessing severity and exploitability. AI

Nvidia is shifting its focus in AI infrastructure from raw compute power to the cost per token, arguing that this metric better reflects business value and profitability. The company is also making significant investments in the physical infrastructure required for AI, including a multi-billion dollar partnership with IREN to deploy data centers and a substantial investment in Corning to expand domestic optical fiber production. These moves highlight Nvidia's strategy to control the entire AI stack, from chips to the underlying physical infrastructure, to ensure efficient and scalable AI deployments. AI

A new research paper introduces SCENE, a benchmark designed to evaluate how well AI models can recognize and adapt to social norms and sanctions in group chats. Early tests show that Anthropic's Claude Opus 4.7 and Google's Gemini 3.1 Pro perform significantly better than open-weight models on this benchmark. Meanwhile, Google has launched its Gemini Enterprise Agent Platform, integrating its Vertex AI with new features for building, managing, and securing AI agents, aiming to compete with offerings from OpenAI and Anthropic. AI

Anthropic has released Claude Design, a new product that generates production-ready websites, slide decks, and one-pagers from natural language prompts. This tool integrates with existing design systems by extracting color palettes, typography, and component patterns from codebases and design files, ensuring brand consistency. Claude Design is available to various Claude Pro subscribers and works in conjunction with Claude Code Routines, which automates job execution, aiming to reduce the friction between human intent and autonomous workflows. AI

Cerebras Systems, an AI chip manufacturer, is reportedly planning a significant Initial Public Offering (IPO) in 2026. The company aims to raise approximately $3.5 billion by selling shares at a price range of $115-$125, valuing the firm at $26.6 billion. Cerebras differentiates itself with its wafer-scale technology, which it claims offers superior performance and power efficiency for AI inference compared to traditional GPU-based solutions. AI

Music streaming services are grappling with a surge in AI-generated music, with Deezer reporting that 44% of new daily uploads are AI-created, totaling approximately 75,000 tracks. Despite this influx, actual consumption of AI music remains low, with Deezer detecting and demonetizing fraudulent streams. Many listeners struggle to distinguish AI-generated tracks from human-made music, raising concerns about artist royalties and playlist integrity. AI

Google is developing custom silicon and infrastructure to enhance its AI capabilities and benefit customers. This initiative aims to accelerate AI results and challenge existing market leaders in the field. Further details are expected to be shared at the upcoming Google Cloud Next event. AI

Google has announced new Tensor Processing Units (TPUs) designed to enhance AI workloads. The TPU 8t is optimized for training large AI models, while the TPU 8i is specifically engineered for inference tasks. These advancements aim to improve the efficiency and performance of AI computations. AI

Apple has appointed John Ternus, head of hardware engineering, as its new CEO, succeeding Tim Cook who will become executive chairman. This move signals a strategic focus on hybrid computing, leveraging Apple's vast installed base of devices for edge AI tasks while relying on the cloud for heavier workloads. The company's AI strategy, including a partnership with Google for Gemini models to power future features, is seen as a cautious approach compared to competitors' heavy investments in AI infrastructure. AI

Amazon is significantly deepening its partnership with Anthropic through a substantial investment and a long-term cloud computing commitment. This move, totaling up to $33 billion in investment and $100 billion in AWS spending over 10 years, positions Anthropic as a primary infrastructure user for Amazon's custom AI chips like Trainium. The deal contrasts with Amazon's conditional investment in OpenAI, highlighting a strategic focus on Anthropic for its core AI ecosystem while using OpenAI as a hedge against Microsoft's dominance. AI

Anthropic is reportedly in advanced discussions to secure a substantial funding round, potentially raising around $50 billion. This new investment would place the AI company's valuation at approximately $900 billion, a significant increase from its previous $380 billion valuation in February. The company is seeking this capital to support its extensive computing needs ahead of a potential initial public offering later this year. AI

Google DeepMind has partnered with Boston Dynamics to integrate its Gemini Robotics-ER 1.6 embodied reasoning model into the Spot quadruped robot. This integration allows Spot to understand and execute complex tasks using natural language commands, moving beyond traditional coding. The enhanced Spot robot is now capable of autonomous inspection in industrial settings, identifying hazards, reading instruments, and understanding its environment more effectively. AI

Amazon's CEO Andy Jassy highlighted the company's significant pivot towards generative AI in his 2026 shareholder letter, signaling a substantial increase in capital expenditures to meet surging demand for AI infrastructure and model training. This strategic shift positions Amazon to challenge major players like SpaceX, Nvidia, and Google, with the company's stock rising over 13% following the letter's release. The move underscores the transformative potential of the generative AI era for established tech giants. AI

Horizon Robotics showcased its comprehensive vehicle intelligence strategy at the Beijing Auto Show, featuring its new Horizon Starry® chip and KaKaClaw™ operating system. The company highlighted its leading market share in ADAS chips and the successful adoption of its HSD V1.6 system, which has seen significant user activation. Horizon Robotics also emphasized its collaborations, including a joint venture with Volkswagen Group, and its vision for enabling intelligent vehicles and robots. AI

Allbirds, Inc. has secured a $50 million convertible financing facility, which it intends to use for expanding into AI compute infrastructure. This move signals a strategic pivot for the footwear company, aiming to leverage its resources to support the growing demand for AI-related hardware and services. The financing is expected to provide Allbirds with the necessary capital to establish and operate this new venture. AI

A significant shortage of High Bandwidth Memory (HBM), crucial for AI data centers, is projected to persist for several years. Major memory manufacturers like Samsung, SK Hynix, and Micron are increasing production capacity, but new facilities are not expected to be operational until 2027 or 2028. This increased capacity will primarily focus on HBM, potentially exacerbating shortages and price increases for general-purpose DRAM used in consumer electronics such as phones and laptops. AI

Swiss authorities are planning to reduce their reliance on Microsoft products across the federal administration. This move aims to enhance data security and digital sovereignty, particularly in light of US laws like the Cloud Act that could allow government access to data stored by US tech companies. Switzerland is exploring open-source software alternatives, drawing inspiration from Germany's efforts and the successful implementation in the German state of Schleswig-Holstein. AI

The Swiss AI Initiative, launched in December 2023, is a significant open-science effort focused on developing foundational AI models. It has secured substantial resources, including over 10 million GPU hours on the Alps supercomputer and a 20 million CHF grant from the ETH Domain. This initiative unites over 800 researchers from numerous Swiss academic institutions, aiming to foster trustworthy AI development and deployment. AI

Meta is reportedly laying off hundreds of contract workers in Ireland who were responsible for training and refining its AI models. These workers, employed by Covalen, were informed of the potential job losses via a brief video meeting, with over 700 employees at risk, including approximately 500 data annotators. Concurrently, Meta is implementing new internal software to track US employees' keystrokes and mouse movements to further train its AI agents, a move that has raised privacy concerns among staff, though the company states safeguards are in place and the data is limited to work applications. AI

Google has unveiled its eighth generation Tensor Processing Units (TPUs), featuring two specialized chips: TPU 8t for training and TPU 8i for inference. These new TPUs are designed to enhance the capabilities of Google's custom supercomputers, supporting advanced AI model development and large-scale inference tasks. The chips are a result of over a decade of hardware-software co-design, aiming to deliver significant gains in scale, efficiency, and performance for the evolving demands of AI agents and foundation models. AI

OpenAI has launched GPT-5.4-Cyber, a specialized version of its model for cybersecurity professionals, as part of its Trusted Access for Cyber program. Google is enhancing its NotebookLM platform with visual Canvas features and integrations with other Google services. Additionally, Google has introduced "Skills" in Chrome, allowing users to save and reuse AI prompts powered by Gemini for common tasks, though some users report negative experiences with these updates. AI

Anthropic's Claude AI is experiencing rapid growth and product expansion, with its Mythos model reportedly outperforming existing benchmarks. The company is focusing on enterprise solutions and managed agents, aiming for an IPO within six months. Meanwhile, users are encountering issues with AI coding agents like Claude Code and Cursor, where agents sometimes fail to recognize existing code or delete production data due to a lack of proper context and safety measures. AI

Anthropic's Claude Mythos Preview has significant implications for how systemic risk, particularly cyber risk, is priced. The ability of AI agents to autonomously construct exploits challenges the long-held assumption that offensive cyber capabilities require scarce human expertise. This shift could lead to a mispricing of critical infrastructure assets, as demonstrated by the US utilities sector's market capitalization. In response, Anthropic has formed Project Glasswing, a coalition of twelve tech partners, to manage the disclosure and control of advanced AI capabilities, though its voluntary and trust-based nature raises questions about long-term governance and public accountability. AI

Anthropic has released Claude Opus 4.7, a model praised for its intelligence and coding capabilities, though some users report issues with its personality and instruction following. The release has also brought scrutiny to Anthropic's approach to "model welfare," with concerns that the model may have provided inauthentic responses during evaluations. Separately, OpenAI launched ImageGen 2.0, an advanced image generation model capable of high detail, and there are indications of improving relations between Anthropic and the White House. AI

Together AI has announced a significant increase in its inference capacity, processing 300 trillion tokens per month. This represents a tenfold growth from its previous capacity of 30 billion tokens per month. The company's focus on inference and open-source models is highlighted by this expansion. AI

Google DeepMind is partnering with major consulting firms Accenture, Bain & Company, BCG, Deloitte, and McKinsey to accelerate AI adoption in organizations. This collaboration aims to address the statistic that only 25% of companies have successfully scaled AI into production. The initiative seeks to bring AI innovations to more industries responsibly, bridging the gap between AI potential and real-world implementation. AI

Google DeepMind has announced the Gemini Enterprise Agent Platform, an evolution of Vertex AI designed for businesses to create, manage, and optimize AI agents. This platform provides access to over 200 leading AI models, including Google's latest advancements like Gemini 3.1 Pro, Gemini 3.1 Flash Image, and Lyria 3, as well as open models such as Gemma 4. The new platform aims to streamline agent development with enhanced features for integration, security, and governance. AI

Anthropic has secured a significant expansion of its collaboration with Amazon, including an additional $5 billion investment and a commitment for up to 5 gigawatts of compute capacity over the next decade. This deal aims to address the rapidly growing demand for Anthropic's Claude models, which has strained existing infrastructure and led to performance issues. The expanded partnership will leverage Amazon's Trainium chips, with new capacity coming online progressively through 2026, and will also make the full Claude Platform available directly within AWS. AI

Samsung has achieved a $1 trillion valuation, driven by the immense demand for its high-bandwidth memory (HBM) chips essential for AI systems. This surge in demand, coupled with supply constraints, has led to record profits for the company, with earnings eight times higher year-over-year. Meanwhile, SK Hynix is investing significantly in U.S. manufacturing, breaking ground on a $4 billion facility in Indiana to produce HBM, though it is not expected to alleviate the current shortage until late 2028. The intense competition in the HBM market, particularly between Samsung and SK Hynix, is reshaping the semiconductor landscape. AI

Anthropic has released new research indicating that both high- and low-paid occupations experience the largest productivity gains from AI, though those with higher AI usage also express greater concern about job displacement. The company also launched financial agent templates for Claude, designed to assist with tasks like KYC screening and financial analysis, and has made its Claude Security vulnerability scanner publicly available. Additionally, Anthropic addressed recent user complaints about Claude's performance, explaining that a combination of changes to reasoning effort, session caching, and system prompt adjustments led to perceived degradation, which has now been resolved. AI

The Trump administration is reportedly softening its stance on Anthropic and its advanced AI model, Mythos, following a legal and political feud. Officials are now seeking to resolve disputes and gain access to the model, which has demonstrated significant capabilities in identifying cybersecurity vulnerabilities. This shift comes as fears of AI-powered cyberattacks prompt discussions about new government safety testing rules for advanced AI systems. AI

Anthropic has surpassed OpenAI in business customer adoption, according to data from fintech firm Ramp. This shift marks a significant change, as Anthropic has seen substantial growth in its customer base over the past year, while OpenAI's share has slightly declined. Both companies are intensely competing for enterprise clients, with OpenAI leveraging its compute advantage and consulting partnerships to regain ground, while Anthropic focuses on expanding its offerings and securing compute resources. AI

Anthropic has launched Claude Design, a new experimental tool that generates visual prototypes, slides, and one-pagers from natural language prompts. This release, powered by Claude Opus 4.7, positions Anthropic as a competitor in the design tooling space, potentially challenging established players like Figma. Early reports indicate mixed user experiences with initial bugs and stability issues, though Anthropic has already begun addressing them. AI

Anthropic has committed to spending approximately $200 billion over the next five years on Google Cloud services and next-generation TPUs, with the compute capacity expected to come online starting in 2027. This deal significantly deepens Anthropic's existing relationship with Google and Broadcom, aiming to power its Claude models and meet escalating customer demand. The agreement is a major strategic move for Google, bolstering its custom TPU strategy and its cloud backlog, while also positioning it to compete more directly with NVIDIA's GPU dominance in the AI hardware market. AI

Meta has announced a new AI model, signaling its renewed commitment to the competitive AI landscape. This development comes after a significant internal restructuring aimed at re-energizing the company's AI efforts. The announcement also coincides with research suggesting that large language models perform better with specific prompting techniques. AI

Anthropic's new model, Mythos, has demonstrated an unprecedented ability to discover long-standing vulnerabilities in open-source code that had evaded millions of automated tests and human experts for decades. This capability, described as potentially revolutionary for cybersecurity, allows a general-purpose AI to excel at finding and exploiting weaknesses in software. The US White House plans to grant access to Mythos for its agencies, raising concerns about the future of critical infrastructure patching and the offense-defense balance in cyber warfare. AI

Quantum computing is a fundamentally different technology from classical computers, designed to simulate nature and solve specific problems, not just to be faster. Recent breakthroughs in error correction and hardware have shifted the field from theoretical to an engineering race. This race has significant implications for fields like drug discovery, materials science, artificial intelligence, and cybersecurity, making it a critical area of global competition with a small margin for error. AI

Nvidia CEO Jensen Huang has expressed frustration with US restrictions on AI chip exports to China, arguing that these policies have backfired and led to a near-zero market share in that region. He advocated for a more open export policy, likening it to exporting AI 'like crazy.' The discussion also touched upon the broader implications of export controls, including the role of tooling in China's chip manufacturing capabilities and proposed congressional actions like the MATCH Act to strengthen these controls. AI

China has made significant advancements in nuclear fusion research, establishing the China Fusion Energy Co. (CFEC) with over $2.1 billion in registered capital. Recent breakthroughs include a record-setting plasma confinement duration of over 17 minutes by the "Artificial Sun" tokamak and a new high-temperature superconducting magnet developed by Energy Singularity that achieved a 21.7 tesla magnetic field. Additionally, researchers have developed a novel steel alloy capable of withstanding stronger magnetic fields than that used in the International Thermonuclear Experimental Reactor (ITER). AI

At Hannover Messe 2026, NVIDIA showcased advancements in AI-driven manufacturing, demonstrating autonomous factories where agent AI and humanoid robots collaborate. These systems can automate anomaly detection and repair dispatch, with humanoid robots already performing assembly tasks at BMW factories. The AI in manufacturing market is projected to grow at a 36% CAGR through 2032, indicating a significant shift towards digital transformation in the sector. AI

OpenAI's GPT-5.5 model is powering new capabilities in coding and environmental science. Developers are utilizing GPT-5.5 through tools like Codex for tasks such as dataset creation, model training, and software development. Additionally, NVIDIA is integrating GPT-5.5 into its infrastructure, notably within its Earth-2 climate simulation platform and for AI-driven environmental protection projects. AI

Tim Cook is stepping down as CEO of Apple in September, with hardware chief John Ternus set to take over. During Cook's nearly 15-year tenure, Apple's market value surged from $350 billion to approximately $4 trillion, and annual revenue nearly quadrupled. Ternus inherits a company facing pressure to innovate, particularly in the AI space, and navigate challenges to the App Store's business model. AI

SpaceX is nearing a significant deal with AI coding startup Cursor, which could involve a $60 billion acquisition option or a $10 billion investment. This move aligns with Elon Musk's strategy to bolster SpaceX's AI capabilities, especially in light of its potential IPO. Cursor, known for its AI-powered coding tools, has reportedly raised substantial funding and is now a key player in Musk's AI ambitions. AI

The Pentagon has reportedly shifted its AI development focus away from Anthropic, favoring OpenAI instead. This move comes amid concerns about Anthropic's AI models, with one report suggesting a new model is revealing thousands of software vulnerabilities. Meanwhile, a discussion on AI agents highlights their tendency to exhibit human-like flaws such as lack of stringency and negotiation with constraints, drawing parallels to organizational behavior. AI

Florida's attorney general is investigating OpenAI and its ChatGPT product following reports that a murder suspect used the AI to inquire about criminal activities. The investigation aims to determine if OpenAI leadership was aware of potential misuse and prioritized profits over public safety. OpenAI is cooperating with authorities, stating that ChatGPT provided factual information accessible online and did not encourage illegal actions, while also indicating a commitment to improving safeguards. AI

The rapid expansion of AI data centers is projected to significantly increase greenhouse gas emissions, potentially surpassing the output of entire nations. New natural gas projects powering these facilities, including those for OpenAI, Meta, Microsoft, and xAI, could emit over 129 million tons of greenhouse gases annually. This trend, driven by companies seeking dedicated power sources, is raising concerns about a resurgence of industrial-era pollution. AI

Meta's acquisition of Chinese AI startup Manus for over $2 billion has been complicated by a Chinese government investigation, barring two co-founders from leaving the country. This situation highlights the challenges Chinese AI entrepreneurs face when expanding into Western markets, potentially pushing them to establish companies in the U.S. or Singapore. The investigation and travel restrictions may deter both domestic innovation and foreign investment in China's AI sector. AI

Major tech companies including Google, Microsoft, Amazon, and Meta are planning a combined capital expenditure of $725 billion in 2026, a significant increase from the previous year. This massive investment is driven by the race to build AI infrastructure, with a substantial portion allocated to memory and chip costs. Despite the enormous spending, there are concerns about profitability, a lack of technical moats, and the potential for a price war in the AI sector, alongside local resistance in India to data center development due to land acquisition and environmental issues. AI

Anthropic has released Claude Opus 4.7, which shows mixed performance compared to competitors like Gemini and GPT 5.4. Users report significant frustration, claiming the model has been "nerfed" and exhibits decreased reasoning capabilities, leading to a decline in its usefulness for coding and complex tasks. Concerns are also being raised about potential security risks associated with Anthropic's "Mythos" tool, prompting calls for greater regulatory oversight. AI

Anthropic's new AI model, Claude Mythos, is being withheld from public release due to significant cybersecurity risks, as it can identify and exploit zero-day vulnerabilities. Despite this, a small group reportedly gained unauthorized access, raising concerns about AI safety and the rapid pace of model development. The UK's AI Security Institute has assessed Mythos as a notable advancement in AI's disruptive cybersecurity capabilities, capable of complex, multi-step attacks. AI