THE RUNDOWN

Anthropic has opened up voice mode to its more powerful Opus and Sonnet models, after it was previously limited to the faster but less capable Haiku model. That's a meaningful jump — Opus and Sonnet handle nuance, reasoning, and complex conversations far better. If you're building voice-powered AI tools or just want a smarter spoken conversation with Claude, this changes things considerably.

Why it matters for Ghana: Developers on the continent building voice AI products — think local language assistants or accessibility tools — now have access to a much stronger model. Worth exploring what's possible on the AI tools front.

Elon Musk tried to spin an optimistic picture on Tesla's earnings call this week, talking up new robotaxi cities and bigger fleets. But the gap between the promises and what's actually happening on the ground is widening — regulatory friction, safety incidents, and slower-than-expected rollouts are all piling up. Meanwhile, Mobileye CEO Amnon Shashua is stepping back from the day-to-day to move into a chairman role, just as Mobileye doubles down on robotaxi and robotics technology. The self-driving space is very much still in flux.

Patreon is laying off around 93 people — a fifth of its entire staff. CEO Jack Conte says the core creator business is healthy, but the platform needs to restructure around how AI is changing its operations. It's a familiar story: a company that championed human creators now trimming headcount partly because of the very technology it's been cautious about. Tough optics.

Why it matters for Ghana: Many Ghanaian and African creators use Patreon as a primary income stream. Uncertainty at the platform level is worth watching — it's a good reminder to diversify where your audience and revenue live.

AMD has unveiled its Helios AI rack-scale system, a direct challenge to Nvidia's dominance in the AI infrastructure market. Shipments start later this year. Nvidia has had an almost unchallenged run supplying the GPUs powering the global AI boom, but AMD is clearly ready to compete at scale — and more competition generally means better pricing and options for cloud providers.

Why it matters for Ghana: African cloud and AI infrastructure costs are heavily tied to what global providers pay for compute. A genuine AMD-Nvidia rivalry could bring those costs down over time.

A revealing report today highlights how AI safety restrictions from OpenAI and Anthropic are getting in the way of legitimate offensive security research — the kind that finds real vulnerabilities before bad actors do. Researchers say the guardrails are blunt instruments that can't distinguish between malicious intent and professional red-teaming. It's a real tension the AI industry hasn't solved yet, and it has implications for how secure the systems we all use actually are. Read more on the security side of this debate.

A new US bill — the AI Kill Switch Act — would give the Homeland Security secretary the power to order the shutdown of AI systems deemed dangerous or out of control. It sounds dramatic, but the underlying question is serious: who gets to decide when an AI has gone rogue, and what happens when that decision is political rather than technical? The bill is early-stage, but the conversation it's sparking is important.

Meta has quietly dropped out of a major clean energy industry group as it accelerates its investment in natural gas to power its AI data centres. The company has been under pressure to explain how it reconciles massive energy consumption with climate commitments — and this move suggests it's chosen scale over optics, at least for now.

Google has rolled out selfie video verification that can be used for account recovery, age verification, and AI avatar creation. It's convenient, but it also means Google now has biometric video data tied to your account. Privacy advocates are already asking questions about how that footage is stored and for how long — fair questions worth raising before you try it.

ChatGPT Health is now rolling out to US users aged 18 and over, with OpenAI once again making the case that giving an AI your medical records can improve personalised health guidance. The promise is real, but so are the concerns around data sensitivity. This is one to watch as it expands globally.

Why it matters for Ghana: If ChatGPT Health eventually reaches African markets, it could be meaningful in contexts where specialist medical access is limited — but data sovereignty questions need answers first.

Meta is now banning users who use its Ray-Ban smart glasses to secretly record and harass people — something that became a notable controversy last year when students demonstrated how easy it was to identify strangers using the glasses and facial recognition. It's a necessary step, though critics will note it took public pressure to get here.

That's your July 24 briefing. A lot moving at once — AI voice quality is genuinely improving, the robotaxi dream is running into real-world friction, and the politics of AI governance are heating up fast. Stay sharp.