Edition 9 · 17 June 2026 · 3 min read

Dependency, vulnerability, and who controls the tools

This edition looks at what happens when the AI tools South Africans and their institutions rely on are controlled, disrupted, or left ungoverned by others. A US government order that cut off access to Anthropic's most capable models over a weekend, and a Chinese regulatory intervention that unwound Meta's deal with an AI startup, both show how geopolitical decisions made elsewhere can reach directly into local businesses, research, and everyday platforms. A critical security flaw in Microsoft's AI assistant, now patched, is a reminder that the same tools carry real risk when vulnerabilities go unaddressed. Closer to home, South African newsrooms are using AI without the policies needed to govern it – a gap that matters especially for organisations whose job is to help the public make sense of what is happening.

Policy & governance

  1. Trump’s Anthropic shutdown just made the case for non-American AI

    The Verge — AIPolicy

    The US government ordered Anthropic, the American company behind the Claude family of AI models, to take its two most advanced models offline over a weekend, blocking access for all foreign nationals including Anthropic's own non-American staff, according to The Verge. The action was imposed with little warning or public explanation, and the company said it had no practical choice but to comply. For South Africa, the episode is a concrete illustration of the risk that comes with depending on US-hosted frontier AI: a foreign government can cut off access to tools that researchers, businesses, and institutions here may rely on, with no notice and no recourse.

    Also reported by The Verge — AI, The Verge — AI, TechCentral, The Verge — AI

Business & economy

  1. Meta reportedly unwinds $2B Manus deal after Beijing pressure

    MemeburnBusiness

    Meta has stopped sharing data with Manus AI and separated the two companies' operations, according to Memeburn citing TechCrunch and Bloomberg, after Chinese regulators reportedly ordered the social media giant to undo its roughly $2 billion acquisition of the AI agent startup on national security grounds. Manus builds software that can complete tasks and automate workflows across applications, going beyond simple question-and-answer tools, and Meta had wanted the technology to deepen AI capabilities across WhatsApp, Facebook, and Instagram. For South Africa, where many small businesses rely on those platforms for customer service, marketing, and sales, the case illustrates that the AI features built into everyday tools can depend on cross-border deals subject to geopolitical intervention.

Society & work

  1. SA newsrooms ill-equipped to deal effectively with AI

    Business Day / BusinessLIVESociety

    South African newsrooms are using AI tools without formal policies to govern how journalists apply them, according to an analysis published by Business Day. The gap matters because media organisations play a central role in how the public understands AI and holds institutions to account, yet the report suggests those organisations are themselves unprepared to manage the technology responsibly. The finding points to a skills and governance deficit in a sector whose credibility depends on getting this right.

Technology & infrastructure

  1. Critical Copilot vulnerability allowed hackers to steal 2FA code from users

    Ars Technica — AITechnology

    Microsoft last week patched a flaw it rated as maximally critical in its M365 Copilot product, an AI assistant built into Microsoft's widely used office software suite. Researchers who found the flaw showed how an attacker could use it to extract one-time login codes and other sensitive data from a user's emails, by hiding malicious instructions inside content that Copilot was processing on the user's behalf – a technique known as prompt injection, where the AI cannot tell the difference between legitimate instructions and commands smuggled in through outside material. South African organisations using Microsoft 365, including businesses and government departments, should verify that the patch has been applied, as the vulnerability exposed data held in email accounts to potential theft.