Loading...
News reporting focused on AI and machine learning, covering the companies behind these technologies, their real-world applications, and the ethical concerns they raise. This includes areas like generative AI (large language models, text-to-image and video), speech tech, and predictive analytics.
OpenAI has launched Lockdown Mode, a security feature that blocks data exfiltration during prompt injection attacks. The mode limits outbound network requests and is now rolling out to Free, Go, Plus, Pro, and self-serve ChatGPT Business accounts. It does not prevent injections from appearing in content but cuts off the final stage of data theft.
SpaceX revealed in a regulatory filing that Google will pay $920 million per month for AI computing power, totaling about $30 billion over the contract's duration. The deal grants Google access to 110,000 Nvidia AI chips and comes as SpaceX prepares for a blockbuster initial public offering next week, with a potential valuation exceeding $1.7 trillion.
A developer built Thousand Token Wood for the Build Small Hackathon, a multi-agent simulation where five woodland creatures trade goods, gossip, and hoard using a Qwen2.5-3B model. The article details how engineered scarcity, smarter prompting, and historical market legends brought the tiny economy to life. Key results include 100% valid JSON output, sustained trades, price crashes, and a widening wealth gap.
Florida has become the first U.S. state to file a lawsuit against OpenAI and its CEO Sam Altman personally. The 83-page complaint accuses the company of marketing ChatGPT as safe while exposing minors to dangerous content and driving dependency. Attorney General James Uthmeier seeks billions in penalties and argues the AI chatbot is a defective product and public nuisance.
SpaceX has secured another major compute deal ahead of its IPO, this time with Google. The agreement, revealed in a regulatory filing, sees Google paying $920 million per month from October 2026 through June 2029 for access to approximately 110,000 NVIDIA GPUs and other components. The deal mirrors a similar arrangement between SpaceX and Anthropic announced in May.
The latest episode of The Vergecast explores Nvidia CEO Jensen Huang's vision for AI-powered laptops, Apple's rumored smart glasses, and the first week of the show's new daily format. The hosts discuss whether consumers actually want the AI-heavy PC experience Big Tech is pushing.
Uber blew through its 2026 AI coding budget by April. Microsoft revoked Claude Code licenses. Priceline saw Cursor renewals jump 4-5x. As token consumption skyrockets with agentic AI, the Linux Foundation launches the Tokenomics Foundation to bring cost discipline. Startups and vendors rush to offer tracking tools, but a common language for token economics is still months away.
A new paper systematically evaluates three variants of projection sharing in transformer attention: shared key-value, shared query-key, and a single projection. The authors found that sharing key and value projections performs on par with standard QKV attention while reducing KV cache by 50% with only 3.1% perplexity degradation. Combining this with grouped-query or multi-query attention can cut cache by up to 96.9%, enabling practical on-device inference.
Airbnb CEO Brian Chesky plans to launch a new AI lab, according to Bloomberg and a source familiar with the matter. Chesky, who has advised OpenAI's Sam Altman and helped broker his return to power, is the latest Silicon Valley figure unsatisfied with frontier AI models. The lab's focus may center on user interaction and design, areas Chesky emphasized at Airbnb. He will remain CEO of Airbnb and not lead the lab himself.
Anthropic co-founder Daniela Amodei downplayed concerns about corporate AI spending slowdowns as the company moves toward an IPO. Speaking at the Bloomberg Tech conference, she emphasized that businesses are still early in their AI adoption journey and that public markets offer the capital needed for frontier model development.
Anthropic has released an open-source reference implementation for autonomous vulnerability discovery and remediation using Claude. The framework includes skills for threat modeling, static scanning, triage, patching, and an autonomous pipeline designed for finding C/C++ memory vulnerabilities. It is based on learnings from partnerships with security teams and is intended as a starting point for building custom vulnerability finding pipelines.
A security researcher built a deliberately vulnerable book review app and spent $1,500 testing 16 different large language models on their ability to exploit a common misconfiguration. GPT-5.5 succeeded 7 out of 10 times, while several models like Gemini and Claude hit security guardrails or never found the right attack vector.