#1Anthropic Accidentally Leaks Its Most Powerful Model Yet — "Claude Mythos"
A misconfigured content management system exposed an unpublished Anthropic blog post revealing "Claude Mythos" (internally also called "Capybara"), a model described as a "step change" above the Opus line. Anthropic confirmed the leak was human error and pulled the post, but the cat is out of the bag: internal docs warn Mythos poses *unprecedented cybersecurity risks*, excelling at finding and exploiting software vulnerabilities at speed.
Source: [Fortune](https://fortune.com/2026/03/27/anthropic-data-leak-reveals-powerful-secret-mythos-ai-model/)
---
#2ARC-AGI-3 Drops — Frontier Models Score Near Zero
The new ARC-AGI-3 benchmark, designed to test reasoning on genuinely novel tasks, handed elite models a humbling result: GPT-5.4, Claude Opus 4.6, and Grok 4.2 all scored between 0% and 0.37% on tasks average humans handle with ease. It's a pointed reminder that raw benchmark leaderboards and real-world generalization are very different things.
Source: [Tech Startups / FutureTech](https://blog.tahababa.com/2026/03/march-27-2026-ai-news-roundup-new-tools.html)
---
#3Google Rolls Out "Personal Intelligence" to All Free US Gemini Users
Gemini's Personal Intelligence feature — which lets the AI draw on your Gmail, Calendar, Drive, Photos, and YouTube history — is now live for free-tier US users, not just subscribers. Opt-in and granular app controls are included, but the breadth of data access is raising eyebrows about the privacy tradeoff baked into Google's ecosystem play.
Source: [9to5Google](https://9to5google.com/2026/03/27/gemini-free-personal-intelligence-rollout/)
---
#4Apple Plans to Open Siri to Any AI in iOS 27
Apple is reportedly moving to let users route Siri requests through third-party AI services — Claude, Gemini, or others — via App Store apps, ending ChatGPT's de facto exclusive slot in Apple Intelligence. This signals Apple is pivoting from picking winners to becoming an AI distribution layer, a strategically powerful position if it can execute.
Source: [WBN Morning Brief](https://www.wbn.digital/wbn-morning-brief-march-27-2026-decisions-drive-markets/)
---
#5SoftBank Secures $40B Bridge Loan to Double Down on OpenAI
SoftBank arranged a $40 billion bridge loan through JPMorgan, Goldman Sachs, Mizuho, SMBC, and MUFG to fund further OpenAI investment and general corporate purposes. It's a massive liquidity move that signals SoftBank's all-in posture on the AI bet — and underscores how much financial infrastructure is now being built around a handful of AI labs.
Source: [Tech Startups](https://techstartups.com/2026/03/27/top-tech-news-today-march-27-2026/)
---
#6Legal AI Startup Harvey Hits $11B Valuation with $200M Round
Harvey closed a $200M raise led by GIC and Sequoia, valuing the legal AI company at $11 billion — up from $8B just months ago. With 25,000+ custom agents deployed across 1,300 customers including most of the Am Law 100, Harvey is showing that vertical AI built on top of foundation models can compound fast when it targets high-value, document-heavy industries.
Source: [Bloomberg](https://www.bloomberg.com/news/articles/2026-03-25/legal-ai-startup-harvey-raises-funds-at-11-billion-valuation) / [CNBC](https://www.cnbc.com/2026/03/25/legal-ai-startup-harvey-raises-200-million-at-11-billion-valuation.html)
---
#778 Chatbot Bills Alive Across 27 US States
Six weeks into the 2026 legislative season, state-level AI regulation is accelerating fast: 78 bills targeting chatbots and AI systems are active in 27 states. The patchwork approach mirrors early social media regulation — lots of energy, inconsistent direction, and real compliance headaches for companies deploying AI nationally.
Source: [Transparency Coalition](https://www.transparencycoalition.ai/news/ai-legislative-update-march27-2026)
---
**The Big Picture**
Today's theme is *exposure* — in every sense. Anthropic's Mythos leak stripped away the curtain on what "the next tier" of AI looks like, and it's apparently powerful enough that even the company building it felt it needed a cybersecurity warning label. Meanwhile, Google is opening its AI to your entire digital life, Apple is commoditizing the AI interface layer, and regulators at 27 state capitals are scrambling to get ahead of systems none of them have tested. The frontier is moving faster than the governance frameworks meant to contain it, and the Mythos leak is a preview of what happens when that gap widens further.
