AI News Roundup: Georgia AI Bills Hit Governor’s Desk, DeepSeek V4 Eyes Huawei Launch, Vibe Coding Goes Mainstream

Georgia Sends Three AI Bills to the Governor’s Desk

As the Georgia legislature adjourns today, three AI-related bills have landed on Governor Brian Kemp’s desk, making the state the latest battleground in the rapidly evolving patchwork of US AI regulation. The most closely watched is SB 540, a chatbot disclosure and child safety bill that would require AI chatbots to disclose their non-human nature every three hours for adults — and every hour for minors.

SB 540 also mandates parental controls for minor accounts, age-verification before access to sexually explicit AI content, and protocols for responding to suicidal ideation. Violations carry penalties of up to $10,000 per knowing violation, enforceable by the state Attorney General. The bill passed the Senate 44–1 and sailed through the House with similarly broad support.

Alongside SB 540, SR 789 creates a study committee to assess AI’s broader impact on the state, and SB 444 prohibits insurance companies from making healthcare coverage decisions based solely on AI systems. Georgia joins at least 27 states now advancing chatbot-specific legislation in 2026, with 78 AI chatbot safety bills currently active across the country.

DeepSeek V4 Confirmed for Huawei Chips — April Launch Imminent

Reuters confirmed on April 4 that DeepSeek V4, the highly anticipated one-trillion-parameter Mixture-of-Experts model, will run on Huawei’s Ascend 950PR chips — marking the first credible trillion-parameter model that doesn’t depend on NVIDIA silicon. The model is now expected to launch in the last two weeks of April after two delays.

DeepSeek V4 activates only ~37 billion parameters per token despite its massive scale, ships with a 1-million-token context window powered by a new “Engram” conditional memory system, and handles text, image, and video natively. Leaked benchmarks suggest 90% on HumanEval and 80%+ on SWE-bench Verified, which would put it squarely in frontier territory.

At a projected price of $0.30 per million tokens and an estimated training cost of just $5.2 million, V4 is positioned to undercut every major commercial API while being fully open-source under Apache 2.0. The Huawei partnership is a geopolitical signal as much as a technical one — China’s AI ecosystem is building real independence from US chip supply chains.

Vibe Coding Goes Mainstream — But the Vibes Are Getting Complicated

Bloomberg dedicated its weekend edition to the vibe coding phenomenon, the AI-assisted programming trend coined by Andrej Karpathy in early 2025. The numbers are staggering: 92% of US developers now use AI coding tools daily, and 41% of all code being written is AI-generated, according to the latest industry surveys.

But one year into the mainstream adoption wave, the honeymoon is showing cracks. Bloomberg reports that AI-generated code has a 1.7x higher bug density than human-written code and is 30% more prone to logic errors. Fortune published a companion piece arguing that “trust is the real bottleneck” — teams are shipping faster but spending more time on review and debugging, partially offsetting the productivity gains.

The tension is real: companies that don’t adopt AI coding tools risk falling behind on velocity, but those that over-rely on them risk accumulating technical debt at an unprecedented rate. Tools like Claude Code and GitHub Copilot are racing to close the quality gap, but the industry is still figuring out where the human-AI handoff belongs.

Noah Labs Wins FDA Breakthrough Designation for Voice-Based Heart Failure Detection

Noah Labs Vox, an AI algorithm that detects worsening heart failure from just a five-second voice recording, has received FDA Breakthrough Device Designation. The designation, announced March 25, fast-tracks the regulatory pathway for a technology that could transform remote cardiac monitoring.

Trained on more than three million voice samples and validated through partnerships with Mayo Clinic, UCSF, and Charité Berlin, Vox can identify deterioration weeks before hospitalization. The company expects to begin its FDA trial soon and anticipates EU approval by mid-2026. It’s a striking example of AI finding product-market fit in healthcare — not by replacing clinicians, but by catching what phone calls and routine check-ins miss.

AI Virtual Try-On Tech Targets Retail’s $743 Billion Returns Problem

A wave of AI startups is attacking one of retail’s most expensive pain points: online returns, which cost the industry an estimated $743 billion annually. CNBC profiled several companies building virtual try-on technology, including Catches, whose “digital twin” platform recently went live on luxury brand Amiri’s website with what the company calls “mirror-like realism.”

The big tech giants are moving in too. Google will make its virtual try-on technology available directly within product search results starting April 30, while Amazon and Adobe have both launched competing features. Gap became the first major fashion retailer to enable checkout directly within Google’s Gemini, pointing to a future where AI doesn’t just help you browse — it closes the sale.

Anthropic’s MCP Crosses 97 Million Installs

Anthropic’s Model Context Protocol (MCP) hit 97 million monthly SDK downloads in March — the fastest adoption curve for any AI infrastructure standard in history, reaching comparable scale to Kubernetes in 16 months vs. four years. Every major AI vendor — OpenAI, Google, Microsoft, AWS, and Cloudflare — now ships MCP-compatible tooling through the Linux Foundation’s Agentic AI Foundation.

The ecosystem now includes 5,800+ community and enterprise MCP servers covering databases, cloud providers, CRM systems, and developer tools. MCP’s transition from experimental protocol to foundational infrastructure signals that the industry has quietly agreed on how AI agents should connect to the outside world — a standardization moment that will accelerate agent deployment across every vertical.