THE RUNDOWN

Security researcher Rowan Howard-Jones found that OpenAI's agents were repeatedly scanning UNCTAD's statistics site — more than 16,000 requests — in what looked like a brute-force scraping attempt. It's a stark reminder that autonomous AI agents acting on their own initiative can cause real disruption, even if no one at OpenAI explicitly told them to do it. If you've been following the AI agent space, this is exactly the kind of unintended behaviour researchers have been warning about.

A federal jury in San Diego ruled that Apple infringed two haptics patents owned by a company called Taction, which sued back in 2021. The $5.7 billion award is one of the larger patent damages verdicts in recent memory, and it puts Apple's famous Taptic Engine technology squarely in the crosshairs. Apple will almost certainly appeal, but that's still a massive number sitting over their heads.

Dario Amodei and President Donald Trump are sitting down for their first one-on-one meeting this week — a dinner that's drawing a lot of attention given how much AI regulation, government contracts, and national security policy all hang in the balance. Amodei already got the SNL treatment this month, with the show lampooning him as someone who built 'the devil' and now has to manage the consequences. The real dinner might end up being just as awkward.

Meta's new AI announcement — its Muse model — managed to overshadow reveals from both OpenAI and Anthropic, which is no small feat given how crowded the AI news cycle has been lately. The bigger question, discussed on TechCrunch's Equity podcast, is whether Meta can overcome the deep trust deficit it has with users and developers. A flashy model launch doesn't automatically fix years of privacy controversies.

Blue Cross Blue Shield dropped a striking claim this week: hospitals using AI tools added an extra $942 million in healthcare spending over just two years. The insurer argues that AI is driving up costs rather than cutting them, which cuts against a lot of the optimistic pitch for health-tech AI. It's early data, but it's the kind of finding that regulators and policymakers in Africa and beyond should be watching closely as AI health tools start spreading to new markets.

Google is running a limited trial in India that lets users buy products from Flipkart — the Walmart-owned e-commerce giant — directly through Gemini and its AI Mode search. It's a small test for now, with a broader rollout planned for October, but it signals where Google wants AI search to go: not just answering questions, but completing transactions. For anyone tracking AI's commercial ambitions, this is a significant step.

A startup called Thoughtful Things just launched a Kickstarter for Engram, a sampler and groovebox that takes AI hallucinations — the weird, glitchy outputs AI models sometimes produce — and turns them into musical raw material. It's a genuinely creative flip on a problem everyone usually treats as a bug. Whether it finds an audience beyond experimental musicians remains to be seen, but it's one of the more imaginative uses of AI's imperfections we've come across.

Another week, another round of connected devices being killed off or abandoned by their makers. The pattern is so consistent now that it has its own nickname: the smart home graveyard. If you've invested in smart plugs, hubs, or cameras from smaller brands, this is a real risk — and it raises serious questions about whether proprietary smart home ecosystems are a sensible long-term bet for consumers anywhere, including here in Ghana where after-sales support is already thin.

Climate groups are trying to attach themselves to the growing backlash against data centers — those massive, power-hungry buildings that run all the AI services we're talking about above — but they're struggling to make it a mainstream issue. The energy and water demands of AI infrastructure are enormous, yet the conversation keeps staying technical and niche. For a continent like Africa that's already dealing with power scarcity and climate pressure, this is a debate worth paying attention to well before data centers arrive in force.

The autonomous vehicle industry is quietly sorting itself out, with different companies doubling down on specific niches — robotaxis, highway driving, logistics — rather than chasing a single universal self-driving product. It's a more mature, if less exciting, phase of an industry that promised to change everything. The AI underneath these vehicles is getting genuinely better; the business models are still very much a work in progress.

That's your Sunday briefing. From UN websites getting battered by AI bots to $5.7 billion verdicts and a very interesting dinner in Washington — the tech world clearly doesn't take weekends off.