Google Expands AI Music Tools

Google turns Gemini into a music studio

Gemini can now generate a 30-second song with lyrics from a single prompt.

Google is adding DeepMind’s Lyria 3 model to the Gemini app, letting users describe a track and get music, vocals, and cover art. The feature is in beta and rolling out to users 18 and older in multiple languages, while YouTube’s Dream Track tool is expanding globally with the same model.

This will get used. The important shift is distribution, because putting music generation inside Gemini and YouTube moves it from a niche demo into everyday creator workflow. Google is trying to get ahead of copyright fights by blocking direct artist mimicry and watermarking tracks with SynthID, but style imitation will still test those boundaries.

The guardrails feel practical. Artists are already split on AI music, and labels are suing model makers while platforms sign licensing deals, so this rollout lands in the middle of an unsettled market. Late at night with headphones on, it will be hard for casual listeners to tell what started as a prompt.

If anyone can make a decent track in 30 seconds, what happens to the value of average music?

OpenAI ties itself to India’s largest IT firm

Tata and OpenAI are building up to 1 gigawatt of AI data center capacity in India.

Tata Consultancy Services said it will start with 100 megawatts of AI infrastructure and expand over time, while also helping companies deploy OpenAI tools across industries. The deal follows Infosys announcing a partnership with Anthropic earlier this week, which shows how India’s biggest IT firms are lining up behind U.S. model makers.

This is a clear alignment. OpenAI gets distribution and political cover in one of the fastest-growing tech markets, and Tata gets early access to high-demand models as Indian clients rush to experiment with AI. The real value here is less about raw capacity and more about who controls enterprise relationships when budgets shift.

The competition is tightening. With Reliance talking about $110 billion in AI infrastructure and the government expecting over $200 billion in spending, India is moving from outsourcing hub to compute host. You can almost hear the hum of server racks in Mumbai and Hyderabad as these plans stack up.

If every major IT firm pairs with a different model provider, how long before clients start choosing sides?

Copilot read emails it wasn’t supposed to

For weeks, Copilot summarized emails marked confidential.

Microsoft confirmed a bug allowed Microsoft 365 Copilot Chat to process draft and sent emails even when data loss prevention policies were in place. The issue, tracked as CW1226324, started in January and a fix began rolling out in early February, though Microsoft hasn’t said how many customers were affected.

This is a trust problem. Enterprise buyers were told AI tools would respect existing security labels, and this undercuts that promise in a very direct way. When confidential tags fail quietly inside tools like Word and Outlook, admins lose visibility and executives get nervous fast.

The timing is awkward. The European Parliament just blocked built-in AI features on work devices over similar concerns, which tells you how sensitive this has become in government circles. Picture an IT admin staring at a compliance dashboard late at night, wondering what else slipped through.

If security labels can be bypassed by a bug, how much faith should companies put in AI guardrails?

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.