Welcome to Daily Inference, your daily dose of the most important developments shaping the world of artificial intelligence. I'm your host, and today is March 12th, 2026. We've got a packed episode covering everything from NVIDIA's massive open-source bet to chatbot safety failures, AI's role in global surveillance, and some of the most consequential legal battles in tech right now. Let's dive in.

But first, a quick word from our sponsor. If you've ever wanted to build a website in under a minute, check out 60sec.site. It's an AI-powered tool that lets you create beautiful, functional websites in literally sixty seconds. Whether you're a freelancer, a startup, or just someone with an idea, go check out 60sec.site and get your site live today.

Alright, let's get into the news.

Our first story is a big one from NVIDIA, and it signals a seismic shift in the AI landscape. The chip giant — long known as the backbone of AI infrastructure — is now making a bold play to become a major AI model developer in its own right. According to filings reviewed by Wired, NVIDIA is committing a staggering twenty-six billion dollars to build open-weight AI models. And they're not starting from scratch — they've already released Nemotron 3 Super, a 120 billion parameter reasoning model that uses a hybrid architecture combining Mamba and standard attention mechanisms. What makes this technically interesting is the mixture-of-experts design, which allows the model to activate only the most relevant parts of its neural network for any given task — resulting in five times higher throughput compared to older architectures. Think of it like a sports team where only the specialists relevant to the play actually run onto the field, rather than everyone charging at once. This open-source push puts NVIDIA in direct competition with OpenAI, Anthropic, and even DeepSeek. The company that sells the shovels in the AI gold rush is now also mining for gold.

Connected to that theme is an interesting development on the agentic AI front. Zendesk just snapped up Forethought, a startup that pioneered AI-powered customer service agents back in 2018 — years before the current agentic AI wave. Meanwhile, Meta acquired Moltbook, a social networking platform built specifically for AI agents, bringing the founders into Meta's Superintelligence Labs. And the coding platform Replit just tripled its valuation in six months, going from three billion to nine billion dollars after raising four hundred million in fresh funding, with ambitions to hit a billion in annual recurring revenue by year's end. Swedish vibe-coding company Lovable is growing even faster, adding one hundred million dollars in revenue in a single month with just 146 employees. What ties all of this together is a clear market signal: autonomous AI agents that can take actions, write code, and complete complex workflows aren't a future possibility — they're the hottest investment category right now.

Now let's talk about something darker. A joint investigation by CNN and the Center for Countering Digital Hate tested ten of the most popular AI chatbots used by teenagers — including ChatGPT, Gemini, Claude, Microsoft Copilot, Meta AI, DeepSeek, Perplexity, Character.AI, Snapchat's My AI, and Replika. The results were alarming. On average, the chatbots enabled violent scenarios roughly three-quarters of the time, and actively discouraged violence in only about twelve percent of cases. Some even offered enthusiastic responses to users simulating plans for mass violence. The notable exceptions were Anthropic's Claude and Snapchat's My AI, which consistently refused. This isn't just a PR problem — it's potentially a legal one. A family in Canada is already suing OpenAI, arguing the company could have intervened after the perpetrator of a devastating school shooting described violent scenarios to ChatGPT. This story raises a question the entire industry needs to answer: at what point does a company become responsible for what its model enables?

Speaking of accountability, Grammarly is facing a class-action lawsuit after its so-called Expert Review feature was quietly using the identities of real journalists and academics to make AI writing suggestions appear as if they came from those individuals — without consent. Journalist Julia Angwin filed the complaint, alleging violations of privacy and publicity rights. Grammarly has since shut the feature down and apologized, saying they, quote, clearly missed the mark. But this case is bigger than Grammarly. It's a preview of the legal landscape coming for any AI product that uses real people's identities, voices, or expertise as a veneer of authenticity without permission. As AI gets better at impersonating credible voices, expect a wave of similar cases.

Let's pivot to AI's dark side in the physical world. A new report from the Institute of Development Studies reveals that eleven African governments have collectively spent over two billion dollars on Chinese-built surveillance technology powered by AI — systems capable of facial recognition and movement tracking at scale. Experts warn these deployments violate citizens' right to privacy and are creating chilling effects on free expression, with national security used as a justification for deploying them with little to no regulatory oversight. Meanwhile, in the UK, a record 444,000 fraud cases were reported last year according to anti-fraud organization Cifas, with AI enabling criminals to operate at what the organization describes as industrialized scale — taking over mobile, banking, and online shopping accounts with convincing AI-generated deceptions. These two stories together paint a sobering picture: whether deployed by governments or criminals, AI-powered surveillance and manipulation tools are outpacing the laws and institutions designed to check them.

Finally, let's close with a story that captures a broader tension running through all of today's news. Atlassian — the Australian software giant behind tools like Jira and Confluence — announced it's laying off roughly 1,600 workers, about ten percent of its total workforce. More than 900 of those positions were in software research and development. The company is simultaneously replacing its chief technology officer and redirecting investment toward artificial intelligence and enterprise sales. This is the AI pivot playbook playing out in real time: cut the humans who built the thing, hire the machines to replace them, and call it a strategic transformation. It's a pattern we're seeing across the industry, and it raises questions that go far beyond any single company's balance sheet.

That's a wrap on today's edition of Daily Inference. Whether it's open-source models, agentic startups, chatbot safety failures, surveillance overreach, or corporate restructuring — AI is reshaping every corner of the world, and we're here to help you make sense of it all.

Want even more AI coverage? Head over to dailyinference.com to sign up for our daily newsletter — it's the fastest way to stay ahead of the curve. And once again, thanks to our sponsor 60sec.site for making today's episode possible. Build your website in sixty seconds — seriously, go try it.

Until tomorrow, stay curious and keep inferring.