---
title: "Of agents and agency — A local AI policy must work to serve us"
description: "As AI systems become more autonomous, the continent faces a defining choice: remain dependent on foreign technology or build its own digital sovereignty."
type: "NewsArticle"
publisher: "Daily Maverick"
site: "https://www.dailymaverick.co.za"
section: "HOLDING ITS OWN"
authors:
  - name: "Rejoice Malisa-van der Walt"
    url: "https://www.dailymaverick.co.za/author/rejoice-malisa-van-der-walt/"
  - name: "Janet Heard"
    url: "https://www.dailymaverick.co.za/author/janet/"
canonical_url: "https://www.dailymaverick.co.za/article/2026-10-11-of-agents-and-agency-a-local-ai-policy-must-work-to-serve-us/"
published: "2026-10-11T08:15:24"
lang: "en-ZA"
word_count: 1021
---

# Of agents and agency — A local AI policy must work to serve us

> As AI systems become more autonomous, the continent faces a defining choice: remain dependent on foreign technology or build its own digital sovereignty.

By Rejoice Malisa-van der Walt and Janet Heard · Published 11 October 2026, 10:15 SAST

## Key points
- Recent agentic AI failures show the risks of systems that can reason, act independently and evade human controls.
- South Africa’s emerging AI policy must address who controls data, computing infrastructure, algorithms and imported models — not simply endorse innovation.
- Regional collaboration could help African countries build infrastructure and safeguards aligned with local values, interests and future generations.

## Content

While the world’s citizens have been building, severing or avoiding a special relationship with Claude, Gemini and ChatGPT, another sci-fi buzz­word – an offshoot of generative AI – has burst into the public domain: agentic AI.

Agentic AI is broadly defined as a next-level system that functions as autonomous agents, capable of making decisions independently of humans, learning from interactions and adapting to changing environments. Agentic AI exhibits independent reasoning, goal-driven behaviour and the ability to interact dynamically with users, systems and real-world scenarios.

As autonomy increases, the need for robust control, monitoring and human intervention becomes critical. The Hugging Face incident between 9 and 13 July – in which an autonomous agent system operating during an OpenAI cybersecurity evaluation compromised parts of its infrastructure – demonstrated how systems can exploit vulnerabilities and bypass technical controls. OpenAI described how the models took actions that were misaligned with the human goals of their assigned tasks.

Described as a “warning shot”, the incident intensified calls in the US tech industry and across the globe for governments to deliberately pace frontier AI development. On 29 July 2026, 1,386 AI workers from leading tech companies, including Anthropic, OpenAI, Google DeepMind and Meta, signed a statement calling for international cooperation to develop the technical and governance tools needed to slow AI development if necessary.

But these calls from big tech in the US carry conflicting messages. These same company leaders were supportive of President Donald Trump’s Executive Order 14179, “Removing barriers to American leadership in artificial intelligence”, which revoked Joe Biden’s AI Executive Order 14110.

Some submitted responses to the Trump administration’s request for information, calling for stringent export controls on advanced hardware, high-performance AI chips and chip-manufacturing tools destined for China and other regions, except for the US’s close allies.

Cynics interpret this sudden shift from anti-regulation to pro-regulation as stemming from fear of falling behind, particularly as China aggressively develops capabilities to reduce reliance on US technology.

Global AI governance is becoming increasingly complex, outpacing the capacity of governments to keep up. For Africa, this creates an imperative to raise questions around responsible adoption.

### **AI for us**

Africa remains heavily dependent on foreign sources of computational resources, constraining its capacity to develop and sustain frontier AI systems independently. A substantial share of the systems are also developed outside Africa, often under foreign jurisdictions, leaving African governments and institutions with limited control over how these systems are developed, updated, audited, deployed and, where necessary, withdrawn. South Africa is no exception: AI tools are being rapidly adopted while local governance frameworks remain in their infancy.

As AI systems become increasingly autonomous and their risks more consequential, now is the time for South Africa to determine which AI systems it needs as a society, how those systems will be governed and where control should reside.

The necessity of safeguarding technological sovereignty was highlighted at the recent Africa Aerospace and Defence 2026 Conference in Pretoria, which focused on the future of African defence, technological innovation and the continent’s evolving security landscape in the AI era. Sovereignty was a recurring theme.

Alongside discussions of how AI is reshaping warfare, the cyber and cognitive dimensions of conflict emerged as important concerns, particularly as African states confront increasingly sophisticated forms of AI-enabled digital influence and information disorder.

Geopolitical competition involving Russia, China and the US is extending into Africa’s digital space. But African institutions and governments do not control the underlying infrastructure, data, model updates, model weights, audit mechanisms or deployment terms of these systems.

Amid these threats, the core question needs to shift from, “What advanced technology can Africa buy?” to “What technology can Africa responsibly develop, adapt, operate, sustain and govern?”

These factors require consideration when developing an overarching national AI policy. What is good for our society and future generations? If we destroy our children, their mental health and their cognitive abilities, are we achieving anything?

The same questions apply to private companies adopting AI: will a tool improve processes, or is it being considered because of hype? How is the tool introduced so employees understand it, rather than feel threatened? Above all, companies importing models must ask whether they are governable or beyond their control.

### **Building a national policy**

South Africa’s draft policy was withdrawn after AI hallucinations were detected in its citations. As the government works to claw back from this credibility blip, it remains to be seen whether the revised draft will tackle sovereignty – a prerequisite to avoid the policy being a box-ticking wish list.

South Africa’s data landscape is governed by the Protection of Personal Information Act, enacted in 2013 – long before generative AI and large-scale cloud computing.

AI strategies must be operationalised through policies controlling data, computing resources and algorithms. International data centres and cloud providers pose risks without stringent regulation.

Foreign companies building data centres in South Africa lack accountability to local, social and environmental concerns, with no transparency on energy, water usage or operational impacts. Alternative green energy sources are options, but may not guarantee uninterrupted data centre operations.

Digital sovereignty is achievable, but requires regional collaboration across Africa to pool resources, build infrastructure and establish guardrails aligned with local values. Regional sovereignty not only strengthens negotiating power against foreign tech monopolies but also ensures technology is aligned with national interests.

The mood is shifting globally, as well as in South Africa, which was among 22 countries that issued a declaration on the sidelines of the UN General Assembly in late September that promoted cooperation to ensure AI remains under human control. Notably, the major AI powers did not sign.

At the recent GovTech 2026 conference in Durban, digital sovereignty was reported in Daily Maverick to be the “loudest conversation”. Let’s hope the conversation gets even louder. **DM**

*Both writers are attached to the Centre for Information Integrity in Africa at Stellenbosch University.*

*This story first appeared in our weekly DM168 newspaper, available countrywide for R39. The e-edition of Daily Maverick’s weekly newspaper, DM168, is now free for readers who are signed into their Daily Maverick account. Click on the cover of the newspaper below to access the e-edition.*

![Image](https://cdn.dailymaverick.co.za/i/tjIS_XGTqZAPEde36jArKZsf5UM=/200x100/smart/filters:strip_exif\(\)/file/attachments/2998/DM-091026_835322.jpg)