April 2026 Open Source AI Recap 🔥 · Open Source AI Builder's Club

April 2026 Open Source AI Recap 🔥

Big month in open source AI. Here's what caught my eye:

1. IBM dropped Granite 4.1 - An 8B model matching 32B MoE performance. Smaller, faster, cheaper — and it's open source. This is the kind of efficiency gains we're talking about when we say "own your AI."

https://firethering.com/granite-4-1-ibm-open-source-model-family/

2. Zig project says no to AI contributions - The Zig programming language officially banned AI-generated code from their project. Does "open source" mean human-only contributions? This conversation isn't going away.

https://simonwillison.net/2026/Apr/30/zig-anti-ai/

3. PyTorch Lightning malware (Shai-Hulud) - Someone shipped themed malware through the PyTorch Lightning AI training library. Check your dependencies, even the popular ones. Supply chain attacks are getting smarter.

https://semgrep.dev/blog/2026/malicious-dependency-in-pytorch-lightning-used-for-ai-training/

4. Claude Code vs OpenClaw drama - Anthropic's coding tool started refusing requests or charging extra if your commits mention "OpenClaw" (an open source competitor). Paid AI locking out open source. Sound familiar?

https://twitter.com/theo/status/2049645973350363168

5. Fine-tuning can unlock copyrighted content - Research showed that fine-tuning on specific data can make LLMs recall copyrighted material. Alignment whack-a-mole continues.

https://github.com/cauchy221/Alignment-Whack-a-Mole-Code

6. Mike: open-source legal AI launched - Legal AI going open source. Another domain getting the open-source treatment.

https://mikeoss.com/

What was your favorite April AI story?

Drop it below 👇

0 comments