State AI power and local deployment choices
This edition covers two developments that sit at different ends of the AI landscape: a reported partnership between Anthropic and the US National Security Agency to deploy AI for cybersecurity and potentially offensive operations, and Lenovo's pitch to South African businesses for a distributed approach to running AI across their own hardware. The first raises questions about how capabilities developed by well-resourced state actors eventually reach criminal networks, with practical consequences for under-resourced South African organisations. The second is a vendor-led account of how local companies are beginning to make infrastructure choices about where and how AI processing happens. Together, the items reflect a tension that runs through much of the current AI moment: decisions made far from South Africa still shape the environment here, while local organisations are being asked to make their own choices about how to engage with the technology.
Business & economy
Lenovo is turning into a one-stop shop for Hybrid AI deployment
Hypertext (htxt)Business
Lenovo Southern Africa used its annual Accelerate event in Johannesburg to position itself as a full-service provider for what it calls Hybrid AI: an approach that spreads AI processing across a company's branch devices, edge hardware, and central data centres rather than relying on a single point. Dean Wolson, the company's general manager for infrastructure services in Southern Africa, told Hypertext that the model is designed to reduce costs by matching the scale of the hardware to the task, and that a small number of local customers have already begun using it. The article is largely based on Lenovo's own event and executive interviews, so the claims about local deployments and productivity gains reflect the company's account rather than independent verification.
Technology & infrastructure
NSA’s Anthropic Mythos Cyber Deal Sparks Alarm
MemeburnTechnology
Anthropic, the US company behind the Claude family of AI models, has reportedly embedded engineers with the National Security Agency to help deploy Mythos, a specialised AI model built for cybersecurity work, according to reporting by the Financial Times and TechCrunch cited by Memeburn. Public reports indicate the model is being prepared for cyber operations that may include offensive use, though the classified nature of the work means the full scope is not confirmed. Memeburn argues the development has practical relevance for South Africa: AI-assisted techniques developed by well-resourced state agencies tend to spread to criminal networks, and organisations with limited security budgets, such as municipalities, clinics and small businesses, are likely to face faster and more sophisticated attacks before they can build adequate defences.