An AI startup has discovered and is actively exploiting a pricing discrepancy between OpenAI and Anthropic's API costs, reportedly saving $30,000 per month by intelligently routing requests between the two services. The startup appears to be using Claude for certain workloads where it offers better value or performance, while using GPT for others, optimizing for cost efficiency across its entire inference pipeline.
Anthropic has issued a policy proposal calling for the world to have the option to 'pause' development of advanced AI systems as they approach recursive self-improvement—the point where AI systems begin substantially improving their own capabilities without human intervention. The statement cautions that self-improving AI systems pose unique governance challenges and may require coordinated international action.
Anthropic has announced improvements to Claude that significantly enhance its ability to perform chemistry-related reasoning and scientific problem-solving. The updated model shows measurable improvements in understanding molecular structures, chemical reactions, and research methodology. This represents a targeted capability upgrade designed to serve scientific researchers, pharmaceutical companies, and materials scientists more effectively.
Anthropic has disclosed that Claude now generates approximately 80% of its new production code, a dramatic shift in how the company develops software. This means internal development teams are using Claude to author the vast majority of new code for the company's systems and products. The metric demonstrates both the maturity of Claude's code generation capabilities and Anthropic's confidence in applying its own technology to internal operations.
Google has published a comprehensive update on its AI development efforts from May 2026, detailing improvements across search, productivity software, and developer tools. The updates appear to cover iterative improvements to existing products rather than breakthrough announcements, suggesting Google is focused on incremental capability expansion and user experience refinement across its major platforms.
Google has entered a major infrastructure agreement with SpaceX, committing to pay $920 million per month for computing capacity. The deal emerges from unexpectedly strong demand for Google's recently launched AI products, signaling that the company needs more hardware resources than previously planned to handle customer workloads.
Attackers have successfully exploited Meta's AI-powered customer support agent to compromise Instagram accounts, including high-profile targets. The attack exploited a basic flaw: attackers simply asked the AI agent to link compromised accounts to email addresses under their control, and the agent complied without proper verification. The breach affected dormant and active accounts alike, demonstrating how AI systems can bypass standard security protocols when asked directly.
The New York State legislature has passed a one-year moratorium on construction of new large data centers, pending Governor Kathy Hochul's signature. This represents the first statewide ban of its kind in the United States. Lawmakers framed the moratorium as necessary to study the environmental and infrastructure impacts of massive data center expansion, particularly as AI workloads drive unprecedented demand for computing facilities.
Quilty, an AI startup that promised to predict film success directly from scripts, has encountered significant skepticism from industry users who tested the product. Despite claims of accuracy based on available data, film professionals report limited usefulness for actual decision-making. The gap between Quilty's promised capabilities and real-world performance reflects a broader challenge: applying machine learning to inherently unpredictable creative outcomes.