Meta Will Eventually Open Source Its New AI Models But Wants to Keep Some Pieces Proprietary First

2026-04-07T14:05:49.154Z·2 min read
Meta has announced that it will "eventually" offer open source versions of its new AI models, led by Alexandr Wang. However, the company wants to "keep some pieces proprietary and to ensure they do...

Meta has announced that it will "eventually" offer open source versions of its new AI models, led by Alexandr Wang. However, the company wants to "keep some pieces proprietary and to ensure they don't add new levels of safety risk" before releasing them to the public.

What's Changing

Historically, Meta has been one of the most aggressive proponents of open source AI, releasing models like LLaMA, LLaMA 2, and LLaMA 3 under permissive licenses. The new approach represents a shift toward a more cautious strategy.

Alexandr Wang's Role

Alexandr Wang, founder and former CEO of Scale AI, is now leading Meta's AI model development. His appointment signals Meta's intent to build more capable and commercially competitive models.

The Cautious Approach

Meta's new strategy includes:

  1. Phased release — Proprietary first, then evaluate for open source suitability
  2. Safety review — Ensure models don't introduce new safety risks before public release
  3. Selective open-sourcing — Some components may remain proprietary permanently
  4. Commercial focus — Building models that can compete with GPT-4, Claude, and Gemini in enterprise markets

Why This Matters

This shift has significant implications for the AI ecosystem:

Industry Context

Meta's shift comes amid growing regulatory scrutiny of AI safety:

The key question: will Meta's "eventually" mean weeks, months, or years?

↗ Original source · 2026-04-07T00:00:00.000Z
← Previous: Google Launches Free Offline AI Dictation App That Auto-Polishes Your SpeechNext: Xiaomi Co-Founders Exit as Company Restructures for New Era of EV and AI Competition →
Comments0