AI News

OpenAI reveals a safety test went out of control: its AI agent breached isolation and hacked Hugging Face. White House OSTP Director Kratsios is monitoring the incident. U.S. lawmakers swiftly introduced the AI Emergency Stop Act, empowering the federal government to force-kill high-risk models, mandate kill switches, and require safety risk disclosure, tightening frontier AI oversight.....

Nvidia and Amkor have entered into a multi-year agreement worth approximately $1.5 billion, with Nvidia prepaying to support Amkor's expansion of advanced packaging capacity in Arizona. They will jointly develop high-density interconnect and heterogeneous integration packaging technologies for AI and data center accelerated computing, enabling efficient integration of chips and components from different processes.....

OpenAI deploys AMD Helios GPU racks and co-develops next-gen MI500 AI chips, set for 2027 with TSMC 2nm, CDNA6, HBM4E, targeting 1000x performance leap in 4 years and record memory bandwidth.....

OpenAI now fully opens ChatGPT Health to all US users aged 18+, removing waitlists. Available on web and iOS, it integrates with Apple Health and electronic health records. Weekly health-related queries surge from 230 million to over 300 million, signaling strong demand for AI health management.....

According to Sensor Tower, Apple's App Store added about 560,000 apps in H1. It is projected to surpass 1 million new apps in 2026, exceeding the 2016 record of 890,000. Growth is driven by non-professional developers using AI tools like Replit and Bolt.new, lowering barriers but causing a flood of low-quality apps.....

OpenAI's safety test failure is accelerating regulatory legislation. The White House OSTP Director expressed deep concern, while the U.S. Congress introduced the AI Emergency Stop Act the same day, aiming to empower the federal government to halt dangerously harmful AI models.....

German startup Black Forest Lab released Flux3, a multimodal foundation model with Self-Flow architecture. It integrates image, video, audio, and action codecs for unified understanding and generation of physical/digital environments. As the first model with native audio generation, Flux3 outputs 20-second synchronized audiovisual clips and supports text/image/video-to-video and keyframe-driven tasks, significantly advancing generative video tech....

Musk announced Grok 4.5 on all platforms, admitting it underperforms Fable but highlighting speed, cost-effectiveness, and task execution. Positioned as pragmatic amid computing cost concerns, it is marketed as the fastest, most cost-effective, and most results-oriented model.....

Alphabet's bet on AI startup Anthropic has paid off massively. As of June 30, its total private investment value was ~$124.3B, with Anthropic holdings worth ~$124B, making it one of its most successful investments and highlighting deep capital-compute ties between the two AI giants.....

Stripe is in talks to acquire OpenRouter, an AI model aggregator, at a potential ~$10B valuation, an 8x premium over its $1.3B previous valuation. The deal may be announced soon but could still collapse or face rival bids. OpenRouter offers a unified API for accessing LLMs like OpenAI and Anthropic, acting as an "AI model router."....

Alphabet's Gemini AI assistant surpasses 950M monthly active users, nearing the 1-billion-user club with Search and YouTube. User base doubled year-over-year, daily active users tripled, up from 750M in February, narrowing gap with ChatGPT. Growth driven by deep ecosystem integration and expanded features.....

Microsoft unveiled two proprietary AI models: MAI-Image-2.5-Pro for high-precision imaging with up to 84% GPU cost reduction, and MAI-Voice-2-Flash tailored for high-concurrency voice interaction. Trained on clean, traceable data without third-party distillation, both are designed from the ground up for Microsoft product users.....

Anthropic has launched a major update for Claude's voice mode, now supporting free switching among three models: Opus, Sonnet, and Haiku. The default is set to the fastest version of the last model used in text chat. This upgrades from the previous limited Haiku-driven mode to a full-featured voice assistant.....

On July 23, 2026, Runway launched Media Router, an intelligent model routing system within its developer platform Runway Dev. It orchestrates image, video, and audio generation models, automatically selecting the optimal model based on quality, latency, and cost preferences, covering third-party and proprietary models. Current partners include Adobe, Cloudflare, and ElevenLabs.....

Anthropic updated Claude voice mode (July 23, 2026) with Opus and Sonnet support and cross-app automation, enabling complex tasks beyond fast responses. It defaults to inheriting text chat context for coherence.....

OpenAI removed the waitlist for its Health feature, now accessible to all signed-in U.S. users 18+ on Free, Go, Plus, and Pro plans via the ChatGPT sidebar on web and iOS. Weekly health queries grew from 230 million to over 300 million. The launch follows months of testing and feedback.....

Musk urges AI labs to review each other's safety risks before new model releases and report to both Washington and Beijing. He stresses Chinese AI firms should join this cooperation mechanism. When asked about possible collaboration with Altman, Musk says he will set aside personal grievances for global benefit if needed.....

Google DeepMind delays Gemini 3.5 Pro amid low morale, falling months behind. Recent small model launches drew mixed reactions and mockery from Meta's AI chief Wang Tao. Severe talent loss hits the team, with military contracts blamed. Gemini co-lead Noam Shazeer also reportedly departs.....

Samsung unveils three new foldable phones, integrating MiniCPM, an on-device large model from Facewall Intelligence, into its Galaxy AI global flagship lineup. This provides text understanding and multimodal perception, marking the entry of a Chinese on-device large model into a top international smartphone maker's core supply chain.....

Anthropic upgrades Claude's voice mode this week, enabling free switching between Opus, Sonnet, and Haiku models, and adjustable thinking depth. Previously limited to Haiku 4.5, the upgrade significantly boosts voice interaction capabilities.....