THE RUNDOWN

The Wikimedia Foundation, which runs Wikipedia, says it 'cannot rule out' that OpenAI's AI agents were behind a major outage back in May. The foundation described the bots as 'rogue' — crawling the site far beyond normal behaviour and overwhelming its servers. This is becoming a pattern: AI systems trained to fetch information from the web are doing so aggressively and without much consideration for the platforms they're hammering. For a free, volunteer-run knowledge base that much of the world depends on, that's a serious problem worth watching on the news front.

Over the past year, OpenAI, Anthropic, and other labs have announced what they're calling breakthroughs on long-standing mathematical problems — in some cases going beyond what human mathematicians have published. The drama isn't just about whether the AI is right; it's about who gets credit, how results are verified, and what it means for the mathematicians whose careers are built on these problems. It's genuinely unsettled territory, and the AI community doesn't fully agree on how to talk about it yet.

Etched, the AI chip startup that's been making noise about building chips specifically for transformer models, is reportedly fielding investment offers at a valuation north of $40 billion — more than double what it was worth just a couple of months ago. That's a staggering number for a company that hasn't shipped at scale yet, but it tells you everything about how desperate investors are to get into the AI hardware race before Nvidia locks up the entire market.

A startup called Reflection has launched Beam, an open-weight AI model it's pitching directly at enterprises and sovereign nations as a lower-cost, lower-compute alternative to Chinese AI models. The pitch is essentially: build your own 'AI factory' without depending on US hyperscalers or Chinese vendors. For African governments and institutions thinking about digital sovereignty, this kind of offering is worth paying attention to — the idea that you could run capable AI infrastructure on your own terms is exactly what many countries say they want.

TikTok has rolled out a conversational AI shopping assistant with one-click checkout, letting users discover and buy products without ever leaving the app. It's a direct play at turning the app's already powerful product-discovery engine into a full commerce platform. Given how much influence TikTok already has over purchasing decisions — especially among younger audiences across Africa — this move could shift real money if it lands well.

Ghost, founded by a 19-year-old, has raised $11 million for its first product: a $3,499 desktop computer called Core, built specifically to run AI agents that can take actions on your behalf. It's essentially a personal AI workstation — not for browsing the web, but for running autonomous agents locally. At that price point it's not for everyone, but the concept of owning your own AI compute rather than renting it from a cloud provider is an idea that's going to keep coming up in tech circles.

Google appears to be expanding its 'Call for Me' AI feature — which currently helps you deal with businesses — so it can also handle personal calls and messages to friends and family. So yes, Gemini could soon ring your mum to say you're stuck in traffic. It sounds small, but it's actually a significant shift: AI assistants moving from professional tasks into personal relationships is a line a lot of people aren't sure they want crossed.

A healthcare startup called Nolla Health has launched an AI-powered acne treatment service in Utah where users scan their skin and receive an actual prescription — no human doctor required. It's the kind of thing that sounds futuristic until you remember that dermatology access is genuinely limited for millions of people. Whether this model holds up to scrutiny from regulators and the medical community will be one to watch.

The Model Context Protocol — MCP — is emerging as a standard for AI agents to communicate with each other. The problem? Security researchers are raising alarms that trust gaps in the protocol can let malicious prompts spread from one agent to another like a chain reaction. As more businesses quietly deploy AI agent networks, this is the kind of security vulnerability that could cause real damage before most people even know what MCP stands for.

Lucid Motors, the luxury EV maker backed by Saudi Arabia's sovereign wealth fund, has reported its lowest production output in nearly two years — and says it's doing so deliberately, citing weak mass-market demand. For a company that was supposed to challenge Tesla at the top end of the market, this is a rough place to be. The EV shakeout is very much still happening, and not every premium brand is going to survive it.

That's your Tuesday sorted. From invisible watermarks to autonomous shopping agents, the pace isn't letting up — and neither are the questions about who's actually in control of any of it.