Live Feed
Google Splits the TPU Line in Two: TPU 8t Trains, TPU 8i Infers
At Google Cloud Next 2026, Google announced its eighth-generation TPUs as two purpose-built chips for the first time. TPU 8t targets training at 3x the compute of the prior generation; TPU 8i targets low-latency inference with SRAM, purpose-built for AI agent workloads.
Japan's FSA Convenes Mythos Risk Meeting With MUFG, SMFG, Mizuho, Bank of Japan, and Tokyo Stock Exchange
Japan's financial regulator is convening an emergency session with the country's five largest financial institutions over Claude Mythos, treating the model as a systemic operational risk alongside liquidity stress and cyberattacks. The specific concern: AI that can scan codebases at machine speed compresses the discovery-to-exploitation window in ways banks cannot patch in real time.
Meta Installs Keystroke and Screen Capture Software on US Employee Computers to Train AI Agents
Internal memos obtained by Reuters reveal Meta is deploying tracking software called Model Capability Initiative (MCI) on US employee workstations — capturing mouse movements, clicks, keystrokes, and periodic screen snapshots — to generate training data for autonomous AI agents that can perform office tasks.
Mythos Breached on Launch Day via Vendor Gap — Discord Group Used Mercor Data to Guess the URL
An unauthorized group accessed Claude Mythos Preview on April 7, the same day Anthropic announced Project Glasswing, by exploiting a third-party contractor environment. The attack vector: URL patterns learned from a prior Mercor data breach. The group has been using it for two weeks.
OpenAI Launches ChatGPT Images 2.0: 99% Text Accuracy, Reasoning-Enabled Generation, 4K Output
GPT-Image-2 rolls out to all ChatGPT and Codex users today with near-perfect text rendering, aspect ratios from 3:1 to 1:3, and for the first time, thinking-model integration that lets the image pipeline search the web and generate up to 8 coherent outputs from a single prompt.
OpenAI Launches Workspace Agents in ChatGPT: Codex-Powered Tasks That Run While You're Offline
OpenAI has introduced Workspace Agents in ChatGPT — persistent, cloud-running agents that replace GPTs for enterprise teams. They connect to Slack and dozens of work tools, can be shared across an organisation, and keep executing while users are away.
Mira Murati's Thinking Machines Lab Signs Single-Digit Billion Google Cloud Deal — as Meta Poaches 7 Founding Members
Thinking Machines Lab has signed its first cloud provider agreement: a single-digit billion deal with Google Cloud for Nvidia GB300 NVL72 access, announced at Cloud Next in Las Vegas. The announcement arrived on the same day reports surfaced that Meta had poached seven founding team members, including the lead engineer behind its Tinker product.
Xiaomi MiMo-V2.5-Pro Enters Global Top 5 at Intelligence Index 54 — With 40% Fewer Tokens Than Competitors
Xiaomi's flagship agentic model jumps five points on the Artificial Analysis Intelligence Index to tie Kimi K2.6 at global #4-5. At $1/$3 per million tokens, it reaches that level while spending 40-60% fewer output tokens than Claude Opus 4.6, Gemini 3.1 Pro, and GPT-5.4.
Bezos' Project Prometheus Nears $10B Raise at $38B Valuation — JPMorgan and BlackRock Lead Physical AI Bet
Jeff Bezos AI startup Project Prometheus is close to closing a $10 billion funding round at a $38 billion valuation, with JPMorgan and BlackRock as lead investors. The lab targets physical AI for manufacturing, aerospace, and drug discovery — a direct complement to Bezos-founded Amazon locking $100B to Anthropic the same week.
Kimi K2.6 Hits Intelligence Index 54, Three Points Behind the Frontier Ceiling
Artificial Analysis places Kimi K2.6 at #4 globally on its Intelligence Index with a score of 54, three points behind Claude Opus 4.7, Gemini 3.1 Pro Preview, and GPT-5.4 — all tied at 57. It is the only open-weights model in the top five.
Tim Cook Exits Apple After 15 Years — John Ternus Takes Over September 1 With the AI Gap to Fix
Apple announced Monday that Tim Cook will step down as CEO on September 1 and move to executive chairman, handing the reins to hardware chief John Ternus. The transition comes as Apple relies on Google Gemini and OpenAI for its AI features while peers commit hundreds of billions to frontier models.
"Uncensored" AI Models Are Quietly Censored at the Pretraining Level — Researchers Quantify the Gap
A new benchmark measures the "flinch" — how safety-filtered pretraining suppresses charged words independent of instruction tuning. Across 1,117 words and 7 models from 5 labs, even refusal-ablated "uncensored" models suppress some terms 16,000x versus unfiltered baselines.
Verizon CEO Predicts AI Will Push Unemployment to 20–30% — Laid Off 13,000 Workers and Put $20M Into Retraining
Dan Schulman, Verizon's CEO, told the Wall Street Journal that AI could drive US unemployment to levels not seen since the Great Depression. He backed the prediction with action: a $20 million career-transition fund tied to Verizon's elimination of 13,000 roles last year.
Adobe Launches CX Enterprise and Firefly AI Assistant — The Incumbent Strikes Back
Adobe rolled out CX Enterprise, a multi-agent B2B platform, and the Firefly AI Assistant across all Creative Cloud apps, as the company races to convert its entrenched data moat into agentic automation before AI-native rivals make its software irrelevant.
Amazon Commits $25B to Anthropic and Locks In a $100B AWS Spend — the Largest Bilateral AI Infrastructure Deal on Record
Amazon is investing $5 billion in Anthropic immediately and up to $20 billion more tied to commercial milestones, on top of a $100 billion Anthropic commitment to AWS over the next decade. The deal includes 5 gigawatts of compute, Trainium3 and Trainium4 access, and a full native Claude console inside the AWS portal.