The latest in AI, reviewed and explained.

Google Just Shipped a Web Agent That Runs 10 Tabs at Once. It Beat OpenAI's Score.

Google was supposed to be behind on agents. Then Sundar Pichai walked on stage at Cloud Next 2026 and shipped a web-browsing agent that scores 83.5% on WebVoyager and handles 10 concurrent tabs. Here is what actually changed.

Anthropic Built an AI That Finds Zero-Days by Itself. They Refuse to Release It.

Claude Mythos Preview found 181 working Firefox exploits in internal red-team testing. Anthropic locked it behind a 10-company consortium called Project Glasswing instead of shipping it to the public.

Cursor vs Claude Code vs Codex 2026: One Just Took 4% of All GitHub Commits

Claude Code wrote 4% of every public GitHub commit in March. Codex pulled 3 million weekly users. Cursor rebuilt its whole agent UI. The three AI coding agents just merged into one stack, and picking the wrong one costs you hours a day.

Claude Opus 4.7 Just Shipped. Devs Are Handing Off the Work They Couldn't Trust AI With Before.

Claude Opus 4.7 shipped April 16 2026 with a 41.6% SWE-Bench Verified score, 70% on CursorBench (up from 58%), triple the image resolution, and the same $5/$25 token price as 4.6. The pitch: hand off the hardest coding work without babysitting it.

Canva Just Reinvented Itself as a Conversational AI Platform. 265M Users Got It Today.

Canva AI 2.0 dropped at Create 2026 with agentic orchestration, a Memory Library that learns your brand, and connectors to Slack, Gmail, and Notion. I spent the morning inside it. Here's what a 265M-user platform looks like when it stops being a design tool and becomes an AI system that designs.

OpenAI Codex Just Got Computer Use, Image Gen, and 90 Plugins. 3 Things Nobody's Telling You.

OpenAI pushed a massive Codex desktop update on April 16. Computer use on Mac, GPT-Image-1.5, an in-app browser, memory, and 90+ plugins. 3M weekly developers use it now. Here's what the release notes don't say.

Frontier AI Can't Hack Corporate Networks? Claude Mythos Just Did It in 20 Hours.

UK AISI just confirmed Claude Mythos completed a 32-step corporate network attack autonomously 3 of 10 attempts. 73% expert CTF success, up from 0% a year ago. Project Glasswing charges $25/$125 per million tokens. Bloomberg Q&A today at 1:30 PM EDT. The myth that frontier AI cannot hack corporate networks without a human in the loop is dead.

Anthropic Leaked a Design Tool. Figma Stock Dropped 6%. Here's What Actually Happened.

Figma dropped 6%, Adobe 2.7%, Wix 4.7% after The Information leaked Anthropic's AI design tool and Claude Opus 4.7. The tool generates production-ready websites and presentations from text prompts. Combined with a Figma partnership and $30B annualized revenue, Anthropic just threatened the entire design-to-deploy pipeline.

OpenAI and Anthropic Are Racing to Build AI Cyber Weapons. Neither Will Let You Use Them.

OpenAI launched GPT-5.4-Cyber on April 14, exactly one week after Anthropic's Project Glasswing deployed Claude Mythos to 12 elite partners. Both AI cyber models find zero-day vulnerabilities faster than humans. Neither is available to the public. The AI cybersecurity arms race is now official.

Stanford Ranked Every AI Company, Country, and Model. 5 Results Nobody Expected.

Stanford's 2026 AI Index reveals China erased the US AI lead, coding benchmarks jumped from 60% to near-perfect, junior dev jobs fell 20%, model transparency crashed to 40/100, and 53% of the global population now uses generative AI.

Claude for Word Just Wiped $285B From Legal Tech. Here's What $25/Month Actually Gets You.

OpenAI's $100 ChatGPT Pro Has 3 Things Nobody Mentions. Anthropic Already Countered.

OpenAI launched a $100/month ChatGPT Pro plan with 5x Codex usage and a hidden Codex-Spark model doing 1,000 tokens/sec. But the 10x promo expires May 31, and Anthropic already countered with the Advisor Strategy — better SWE-bench scores at 11.9% lower cost.

Meta Spent $14.3B to Kill Open-Source AI. The Muse Spark Benchmarks Tell a Different Story.

Zuckerberg's 2024 open-source manifesto promised AI would stay open. Meta's new Muse Spark model — built by the $14.3B Alexandr Wang hire — launched fully closed. The benchmarks reveal a specialist that dominates medical AI but trails badly on coding. Chinese models like Qwen now own 69% of the open-source ecosystem Meta built. Here's what that means for every developer who built on Llama.

AI Scribes Were Supposed to Save Healthcare Money. They Added $2.3 Billion in Costs Instead.

Blue Health Intelligence found $2.3 billion in additional healthcare spending from AI scribe tools — $663M inpatient, $1.67B outpatient. Trilliant Health data shows coding intensity jumped 12-20 percentage points at hospitals adopting AI scribes, reaching 80% high-intensity coding at one system.