Anthropic, valued at nearly $1 trillion, has published new research revealing findings about how large language models function internally. The research adds to the company's track record of publishing technically rigorous work on AI safety, interpretability, and behavior. While the specific findings require technical context to fully understand, the research contributes to the broader effort to make AI systems more transparent and predictable to their users and developers.
Anthropic is rolling out localized pricing for Claude in India, recognizing the country as its largest market outside the United States. This move makes Claude more accessible to Indian businesses and developers by adjusting prices to reflect local purchasing power and economic conditions. The pricing adjustment reflects Anthropic's strategy to compete aggressively in international markets where demand for AI services is growing rapidly but price sensitivity is higher than in developed Western markets.
Apple has filed a trade secrets lawsuit against OpenAI, alleging the AI company engaged in systematic efforts to access Apple's confidential technology and recruit its employees using improper methods. According to the complaint, OpenAI's hardware head allegedly asked Apple job candidates to bring unreleased product components and samples to interviews, and the lawsuit includes claims about unauthorized system access and employee misconduct ranging from casual to serious.
Google is embedding its Gemini AI assistant into Waze, the navigation and traffic app, to give drivers more personalized control over their trips. The integration includes four new features, with two directly leveraging Gemini's capabilities. Users will be able to interact with the app through AI-powered voice commands and natural language requests, making it easier to adjust routes, find specific types of destinations, and customize the driving experience without manual tapping.
Apple has released the first public beta of iOS 27, which includes significant upgrades to Siri powered by on-device AI improvements. Early testers report that Siri is more responsive, capable of handling complex multi-step requests, and better integrated with iOS features than previous versions. The improvements reflect Apple's focus on making AI features feel natural and useful in everyday device interaction rather than flashy or experimental.
Meta's AI-powered advertising tools, designed to automate and optimize campaign performance, are creating unexpected headaches for brands. The AI systems make autonomous decisions about ad placement, audience targeting, and budget allocation without sufficient transparency or control, leading to ads appearing in inappropriate contexts or reaching unintended audiences. Brands report difficulty understanding why certain ads perform as they do and limited ability to override AI recommendations that conflict with brand safety guidelines.
Meta's capital expenditure on AI infrastructure has exploded, rising from roughly $10 billion to $50 billion in less than two years as the company races to build the computational capacity needed for its AI ambitions. This massive investment is funding the construction of large-scale data centers designed to train and run large language models, search ranking systems, and content recommendation algorithms. The scale of spending is reshaping local communities where Meta is building these facilities, sparking debates over energy usage, real estate impacts, and the concentration of AI power.
Microsoft has developed new methods for verifying Rust cryptographic code in SymCrypt, its cryptographic library, allowing developers to verify code correctness as they write it rather than after deployment. The approach balances security assurance with development speed and maintainability, addressing a critical gap in cryptographic software development. By catching errors early and verifying code against cryptographic standards automatically, the method reduces the risk of subtle bugs that could compromise security.
Security researchers have developed a new defensive technique called "context bombing" that uses prompt injection tactics to trick AI-powered hacking agents into shutting down before they can compromise systems. Rather than trying to prevent all prompt injection attacks, defenders are now deliberately injecting misleading instructions into AI agents used by attackers, effectively disarming them. The technique exploits the vulnerability of AI systems to conflicting or nonsensical instructions embedded in their context.
Researchers have demonstrated a proof-of-concept combining quantum computing and AI to design novel peptides for drug development, particularly targeting rare diseases and underserved populations. The team assembled funding and resources to show how quantum computers can handle the complex molecular optimization problems that AI alone struggles with, creating a complementary tool set for pharmaceutical discovery. The breakthrough, though still in early research stages, opens a path for accelerating drug development timelines and reducing costs for conditions that typically receive limited research investment.