Model releases, frontier-lab strategy, AI policy and the economics of machine intelligence.
OpenAI's Bio Bug Bounty invites vetted security researchers to find universal jailbreaks in GPT-5.5's biosafety guardrails — a rare move that signals proactive AI risk management at the frontier.
Saudi Arabia's Project Transcendence and Abu Dhabi's national AI infrastructure drive represent the most ambitious sovereign computing buildout since national space agencies were established — and they are reshaping where AI is trained, who controls it, and at what cost.
OpenAI's newest release claims the top reasoning and instruction-following slot in its model lineup, arriving as a stepping stone ahead of the full GPT-5 flagship and a direct challenge to Google Gemini 2.5 and Anthropic Claude 3.7.
An initial $10 billion stake, with potential for $30 billion more, cements Google's position as Anthropic's largest institutional backer and signals a new era of hyperscaler-driven AI funding.
A landmark PwC study of 1,217 senior executives across 25 sectors reveals a stark and widening divide between AI leaders generating 7.2x more value and a majority of organisations still stuck in pilot mode.
China's DeepSeek drops V4 Pro and Flash on April 24, pushing the boundaries of open-source AI with a 1 million-token context window, Hybrid Attention Architecture, and pricing that continues to undercut Western rivals.
The latest Gemini model outperforms rivals on mathematical reasoning and long-context tasks, as Google doubles down on AI infrastructure with $75 billion in capital expenditure planned for 2026.
The AI-native search startup has tripled its annual recurring revenue in under 12 months, signing deals with Goldman Sachs, McKinsey, and a dozen Fortune 500 legal teams — raising fresh questions about Google's moat in knowledge work.
The first major cluster of OpenAI's $500 billion Stargate project has come online in Abilene, Texas — consuming enough electricity to power 80,000 homes and signaling a new era in AI infrastructure scale.
The federal government has sided with Elon Musk's xAI in a constitutional challenge against Colorado's AI consumer protection statute, set to take effect June 30.
SoftBank-led round values Sarvam at $1.4B as BharatGPT-3 debuts with 70B parameters covering 22 Indian languages — including seven with no prior commercial LLM coverage — and 14 million monthly active users in live government deployments.
Microsoft confirms full agentic runtime for Copilot Studio ahead of its May developer conference, with 3,200 enterprise customers already logging 18 million autonomous task completions in private preview.
Hangzhou-based DeepSeek ships its next-generation reasoning model at roughly one-twentieth the per-token cost of OpenAI's top tier, reopening the pricing war that Mistral and Meta reignited earlier this spring.
Saudi Arabia, UAE, and Qatar are channeling energy revenues into domestic AI infrastructure — data centers, national language models, and regulatory frameworks designed to make them self-sufficient in the AI era.
OpenAI's latest release goes beyond chat — GPT-5.5 is designed to take on multi-step agentic work with fewer tokens and stronger guardrails, setting the stage for a new competitive round with Anthropic.
Meta's latest open-source model outscores GPT-4.5 on key reasoning tests and is already available for free via the Meta AI app — forcing rival API providers to slash prices within days of launch.
Internal benchmarks circulating ahead of Apple's June developer conference suggest the company's next-generation on-device AI—running entirely on M4 silicon—matches cloud-tier performance across key reasoning and coding tasks.
MIT's Daron Acemoglu, co-winner of the 2024 Nobel Prize in Economics, warns that AI's productivity gains flow primarily to capital owners, not workers — and calls the current trajectory 'a shitshow in the making.'
The Model Context Protocol-based integrations — now available to all Claude users, with mobile in beta — let AI agents read and act inside GitHub, Notion, Google Drive, and dozens of other tools without copy-paste workflows.
Anthropic's latest flagship achieves 92.3% on SWE-Bench Verified — a 7-point jump — with revamped extended thinking and a 200K-token context window that holds fidelity near the limit
The French lab's new flagship lands within striking distance of GPT-5 and Claude 4.5 on benchmarks while pricing roughly 50% below, pulling forward an uncomfortable question for the frontier premium tier
Sundar Pichai's Cloud Next disclosure marks a structural shift: AI-assisted coding has crossed from productivity tool to default engineering workflow at Google scale
Presented at ICLR 2026, the algorithm slashes LLM inference memory costs and opens a path to trillion-token context windows at scale
New Gemini Enterprise Agent Platform and eighth-generation TPUs signal Google's all-in bet on the agentic enterprise
