#1Anthropic Accidentally Leaks "Claude Mythos" — Its Most Dangerous Model Yet
A CMS misconfiguration exposed nearly 3,000 unpublished Anthropic assets, including blog drafts for an unreleased model codenamed "Mythos" (also "Capybara"). Leaked internal docs describe it as a "step change" beyond any existing model and warn it poses "unprecedented cybersecurity risks" by autonomously finding and exploiting software vulnerabilities faster than defenders can respond. Anthropic confirmed the drafts were real but says a general release won't happen until the model is made more cost-efficient.
Sources: [Fortune](https://fortune.com/2026/03/27/anthropic-leaked-ai-mythos-cybersecurity-risk/) | [The Decoder](https://the-decoder.com/anthropic-leak-reveals-new-model-claude-mythos-with-dramatically-higher-scores-on-tests-than-any-previous-model/) | [Gizmodo](https://gizmodo.com/leaked-anthropic-model-presents-unprecedented-cybersecurity-risks-much-to-pentagons-pleasure-2000739088)
---
#2Second Wave of AI Chip Smuggling Charges — Six People Down in 10 Days
The DOJ charged three more individuals (a Hong Kong national and two US citizens) on March 26 for attempting to smuggle $170M in export-controlled Nvidia GPUs to China via Thailand. This follows the March 19 arrest of Supermicro co-founder Wally Liaw, accused of orchestrating a $2.5B scheme involving fake server repackaging with hair dryers to swap serial numbers. Each count carries up to 20 years in prison.
Sources: [The Register](https://www.theregister.com/2026/03/26/gpu_china_smuggling_charges/) | [Washington Times](https://www.washingtontimes.com/news/2026/mar/26/three-men-charged-smuggling-170m-export-controlled-chips-china/) | [CNBC](https://www.cnbc.com/2026/03/19/us-tech-execs-smuggled-nvidia-chips-to-china-prosecutors-say.html)
---
#3Google Launches Gemini 3 Deep Think and Gemini 3.1 Flash Line
Google went aggressive on two fronts this week: Gemini 3 Deep Think is now live for Ultra subscribers, targeting scientific and engineering reasoning rather than everyday use. Separately, Gemini 3.1 Flash-Lite launched at $0.25/million input tokens with 2.5× faster responses, and Flash Live entered developer preview — positioning Google for the agent application wave expected within 30 days.
Sources: [LLM Stats](https://llm-stats.com/ai-news) | [LabLA](https://www.labla.org/latest-ai-model-releases-past-24-hours/ai-model-releases-industry-updates-march-27-28-2026/)
---
#4OpenAI Crosses $25B ARR, Eyes IPO as Soon as Late 2026
OpenAI has surpassed $25 billion in annualized revenue and is reportedly taking preliminary steps toward a public listing. Rival Anthropic is close behind at ~$19B ARR, with paid subscriber counts more than doubling this year. The revenue race is now a full sprint — and the gap between the two is narrowing.
Sources: [LLM Stats](https://llm-stats.com/ai-news) | [Crescendo AI](https://www.crescendo.ai/news/latest-ai-news-and-updates)
---
#5Shopify Builds "Agentic Storefronts" Inside ChatGPT, Gemini, and Copilot
Shopify merchants can now sell directly inside AI assistants — ChatGPT, Google's AI Mode, Microsoft Copilot, and the Gemini app — without users ever leaving the chat interface. It's the most concrete example yet of AI platforms converting from information tools to commercial transaction layers.
Sources: [LabLA](https://www.labla.org/latest-ai-model-releases-past-24-hours/ai-model-releases-industry-updates-march-27-28-2026/)
---
#6Claude Goes Full Computer-Use — Mouse, Clicks, and All
Anthropic expanded Claude's computer-use capability so it can now operate a desktop autonomously: pointing, clicking, scrolling, and navigating any screen when no native integration exists. Claude paid subscribers have more than doubled in 2026, suggesting this kind of agent utility is driving real adoption.
Sources: [LabLA](https://www.labla.org/latest-ai-model-releases-past-24-hours/ai-model-releases-industry-updates-march-27-28-2026/) | [Crescendo AI](https://www.crescendo.ai/news/latest-ai-news-and-updates)
---
#7Congress Moves to Study AI's Job Impact — Bipartisan Commission Proposed
A bipartisan Senate group introduced legislation to create a federal commission studying AI's economic and workforce effects. It's a signal that Washington is shifting from "should we regulate AI?" to "what do we do about mass displacement?" — even if no enforcement teeth are attached yet.
Sources: [Crescendo AI](https://www.crescendo.ai/news/latest-ai-news-and-updates)
---
**Big Picture**
Today's theme is **capability without control**. Anthropic leaked its most powerful — and most dangerous — model by accident, days after the industry's chip-export regime was exposed as riddled with $2.5B holes. Meanwhile OpenAI and Google are racing on revenue and model releases while AI agents quietly colonize commerce and desktops. The pattern is consistent: the technology is moving faster than the institutions — legal, regulatory, and corporate — designed to contain it. The Mythos leak isn't just embarrassing; it's a preview of the coordination problems ahead.
