Cursor is betting that developers are frustrated enough with GitHub to switch platforms entirely. The company, which built its reputation on an AI-powered code editor, announced a new code hosting service yesterday that goes head-to-head with Microsoft’s GitHub.

The timing isn’t subtle. GitHub’s been catching heat from developers over AI training policies, changing terms of service, and the usual Microsoft-flavored product decisions that make people nervous. Cursor is positioning itself as the developer-first alternative, though details on what actually makes the platform different (beyond “we made an editor people like”) are thin.

If you’re already using Cursor’s editor and want tighter integration with your hosting, this matters. If you’re happy with GitHub or already moved to GitLab or Codeberg, you can probably ignore it. The real test will be whether Cursor can offer something compelling beyond “not GitHub” and whether they can handle the operational complexity of running a code hosting platform at scale.

Mojo finally goes open source

Meanwhile, Mojo delivered on a promise that’s been outstanding since May 2023. The programming language released its compiler and toolchain under an Apache 2 license last week, shortly after shipping version 1.0.

For context: Mojo launched with big ambitions to be a Python superset, fast enough to compete with C++ for systems programming while staying compatible with existing Python code. That compatibility promise quietly changed around August 2025. The current version is Python-inspired but not a strict superset, which means you can’t just drop Python code into Mojo and expect it to work.

Modular, the company behind Mojo, now belongs to Qualcomm. The open source release includes the full compiler toolchain, so developers can actually build and contribute to the language instead of just using whatever Modular ships.

Who should care: if you’re working on performance-critical code and want Python-like syntax without Python-like speed penalties, Mojo’s worth looking at now that it’s open. The 1.0 release suggests the language is stable enough for real projects, and the open source release means you’re not locked into Modular’s roadmap.

Who can skip it: if you’re happy with Python’s performance or already committed to Rust/Go/Zig for systems work, nothing here changes your world. The loss of true Python compatibility makes migration harder than originally promised.

OpenAI’s security theater week

OpenAI had a busy PR day yesterday, announcing new safeguards after the Hugging Face breach, a teen-focused ChatGPT product, and an initiative to help government institutions with AI oversight in national security contexts.

The safeguards piece is the most substantive: more detailed monitoring during model development, stronger emphasis on alignment and security during post-training, and what they’re calling “pacing” of model development based on cyber-critical capabilities. Translation: they’re trying to get ahead of the “AI makes hacking too easy” problem before it becomes a Congressional hearing.

ChatGPT for Teens adds parental controls and guardrails designed to prevent cheating and filter harmful content. It’s launching years after teens started using ChatGPT anyway, which makes this feel more like liability management than product innovation.

The national security oversight initiative is OpenAI offering “tools, training, and expertise” to government institutions. No details on what those tools are, what the training covers, or whether this is a product, a partnership, or a PR campaign.

All three announcements smell like getting out in front of regulation. OpenAI’s betting that proactive “look how responsible we are” moves buy them room to keep shipping fast. Whether it works depends on how seriously regulators take the actual implementations versus the blog posts.