AI News Roundup: Anthropic’s Mythos Leak, Reddit’s Bot Crackdown, White House AI Framework

Anthropic’s “Mythos” Model Leaked — and It’s Raising Alarms

The biggest AI story of the past week landed when a misconfigured content management system exposed nearly 3,000 unpublished assets from Anthropic’s blog, including details about a new model called Claude Mythos (internally codenamed Capybara). Anthropic has confirmed Mythos is real, calling it “a step change” in capabilities and “the most capable model we’ve built to date.”

What makes this leak especially consequential is the cybersecurity angle. Leaked internal documents warn that Mythos is “currently far ahead of any other AI model in cyber capabilities” and could “exploit vulnerabilities in ways that far outpace the efforts of defenders.” Anthropic is reportedly briefing top government officials on the risks and restricting early access to organizations focused on cyber defense. The leak itself — caused by human error — adds an uncomfortable irony to the story.

Reddit’s Bot Crackdown Goes Live Today

Starting today, March 31, Reddit is rolling out its most aggressive automation policy yet. Automated accounts will now carry mandatory [App] labels on their profiles, and accounts flagged as potentially non-human will face verification challenges. Reddit says it’s already removing around 100,000 bot accounts daily.

The policy draws a clear line: using AI tools like ChatGPT to draft a comment is fine at the platform level, but fully unattended bot accounts posting autonomously will need to verify or get labeled. Verification methods under consideration include passkeys, third-party biometrics, and government ID services — all designed so Reddit never sees a user’s real identity. Community moderators retain the power to set stricter rules around AI-generated content in their own subreddits.

White House Drops National AI Policy Framework

On March 20, the White House released its National Policy Framework for Artificial Intelligence — the most concrete blueprint yet for how the administration wants Congress to approach AI regulation. The framework explicitly rejects creating a new federal AI regulatory body, instead pushing sector-specific oversight through existing agencies and industry-led standards.

Key provisions include parental controls and age-assurance requirements for AI services accessible by minors, protections against AI-generated child exploitation material, and a controversial stance that training AI on copyrighted material does not violate copyright law (while leaving courts to decide specific cases). The framework also calls for federal preemption of state AI laws — a move that would override efforts like California’s SB 1047. Democrats remain skeptical, and any bipartisan legislative path will be difficult to navigate.

Anthropic Eyes $60 Billion IPO as Early as October

In parallel with the Mythos news, reports emerged that Anthropic is in preliminary IPO discussions with Goldman Sachs, JPMorgan, and Morgan Stanley, targeting a listing as early as October 2026 that could raise over $60 billion. The company’s most recent private valuation hit $380 billion, and annualized revenue has surged past $19 billion — more than double its $9 billion run rate at the end of 2025.

If both Anthropic and OpenAI go public this year as planned, they’d rank among the largest venture-backed IPOs in history, alongside SpaceX. Enterprises now account for roughly 80% of Anthropic’s revenue, with eight Fortune 10 companies among its customers.

Google Translate Gets Gemini-Powered Real-Time Headphone Translation

Google expanded its Live Translate feature to iOS and several new countries including the UK, Japan, and Nigeria. The feature, powered by Gemini AI, delivers real-time translation through any pair of headphones while preserving the speaker’s tone and cadence across more than 70 languages.

Beyond headphones, Google also rolled out Gemini-powered text translations that handle idioms, slang, and context-dependent phrases more naturally. The upgrades are available now in Google Translate for iOS and the web, supporting English and nearly 20 languages.

OpenAI’s Acquisition Spree Continues

OpenAI has already completed six acquisitions in 2026, matching its total for all of 2025 in just three months. The latest pickups include Astral, a maker of open-source developer tools (acquired March 19), and Promptfoo, an open-source AI testing framework. The buying spree signals OpenAI’s strategy of consolidating the developer tooling ecosystem around its platform ahead of its own expected IPO.

EU Council Moves to Streamline AI Act Rules

The European Council agreed on its position to simplify certain provisions of the EU AI Act, including a new prohibition on AI systems that generate non-consensual sexual or intimate content and child sexual abuse material. The move signals that even as the EU tightens rules in targeted areas, there’s growing pressure to reduce compliance burdens on businesses deploying lower-risk AI applications.