AI News Roundup: Anthropic’s Mythos Leak, Social Media Addiction Verdict, White House AI Framework

Anthropic’s “Claude Mythos” Leak Reveals a Step-Change Model

In one of the most consequential AI leaks in recent memory, Anthropic accidentally exposed details of a powerful in-development model called Claude Mythos (codenamed Capybara) after internal documents were left in a publicly accessible, unencrypted data store. Fortune first reported the leak on March 26, and by March 27 the story had ricocheted across finance and tech media.

The leaked materials describe Mythos as a “step change” in reasoning capability — a new tier above Opus that scores dramatically higher on software coding, academic reasoning, and cybersecurity benchmarks. Anthropic confirmed the documents were “early drafts” and said it is working with a small group of early-access customers to test the model. Internal safety assessments warn that Mythos could “significantly heighten cybersecurity risks” by rapidly identifying and exploiting software vulnerabilities.

Bitcoin and software stocks briefly dipped on March 27 as markets digested the cybersecurity implications. Anthropic says it will focus initial deployment on defensive cybersecurity applications and emphasizes that the leak was a CMS misconfiguration, not a breach.

Landmark Verdict: Meta and Google Liable for Social Media Addiction

A Los Angeles jury on March 25 found Meta Platforms and Alphabet liable in the first-ever product-liability trial over youth social media addiction. The plaintiff, identified only as KGM, alleged that Instagram and YouTube were deliberately designed to be addictive — and the jury agreed, awarding $6 million in total damages ($3M compensatory, $3M punitive), with Meta responsible for 70% and YouTube 30%.

The verdict is a watershed moment. It is the first of thousands of similar lawsuits awaiting trial, and commentators are drawing comparisons to the 1990s legal crusade against Big Tobacco. Meta shares dropped on the news, and Bloomberg reported the companies now face “Big Tobacco-like fallout.” Expect this to accelerate both legislative action and corporate policy changes around youth safety features.

White House Unveils National AI Policy Framework

On March 20, the White House released a sweeping National Policy Framework for Artificial Intelligence — its most detailed AI legislative blueprint to date. The central argument: a patchwork of state-level AI laws is hindering innovation and should be preempted by unified federal legislation.

Key proposals include no new federal AI regulator (existing agencies like the FTC and SEC would oversee their sectors), child safety rules requiring age assurance and parental controls on AI platforms, standardized data center permitting, and strong free speech protections limiting government coercion of AI providers. The framework also calls for precluding states from regulating AI model development or imposing liability on developers for third-party misuse.

Congress has so far declined to pass broad federal preemption of state AI laws, and the framework faces a divided legislature. But the White House says it believes the plan can generate bipartisan support and wants it codified “this year.”

Yann LeCun’s AMI Labs Raises $1.03B — Europe’s Largest Seed Round

Turing Award winner Yann LeCun officially left Meta to co-found AMI Labs, which closed a record-shattering $1.03 billion seed round at a $3.5 billion pre-money valuation. The round was co-led by Cathay Innovation, Greycroft, Hiro Capital, HV Capital, and Bezos Expeditions, with participation from Tim Berners-Lee, Mark Cuban, and Eric Schmidt.

AMI is building “world models” based on LeCun’s Joint Embedding Predictive Architecture (JEPA) — AI that learns from reality rather than text alone, targeting industrial, robotic, and healthcare applications. The company is headquartered in Paris with offices in New York, Montreal, and Singapore.

Anthropic Hits $19B Annualized Revenue as OpenAI’s Enterprise Share Falls

Anthropic reached $19 billion in annualized revenue in March — more than double its $9 billion run rate at the end of 2025. Meanwhile, OpenAI’s share of enterprise AI spending fell from 50% to 27% over the past year, while Anthropic’s climbed to 40%. The shift reflects growing enterprise preference for Claude in coding, analysis, and agentic workflows.

The Pentagon’s decision to label Anthropic a supply-chain risk (now temporarily blocked by a federal judge) has not slowed commercial adoption. More than 30 employees from OpenAI and Google DeepMind signed a brief supporting Anthropic in the lawsuit, calling the designation “an improper and arbitrary use of power.”

Model Context Protocol Crosses 97 Million Monthly Downloads

The Model Context Protocol (MCP), originally created by Anthropic and donated to the Agentic AI Foundation under the Linux Foundation, hit 97 million monthly SDK downloads in March — up from roughly 2 million at launch in November 2024. That’s a 4,750% growth rate in 16 months.

The MCP server ecosystem now includes over 5,800 community and enterprise servers, and the protocol has been adopted by every major AI provider: Anthropic, OpenAI, Google, Microsoft, and Amazon. Analysts say MCP has transitioned from experimental standard to foundational agentic infrastructure.