AI for Individual Rights Newsletter #8

Welcome to the eighth edition of the Human Rights Foundation's AI for Individual Rights newsletter.

We begin with research from the African Digital Rights Network, which documents how 11 African countries have collectively spent more than $2 billion on Chinese-built “smart city” AI-powered surveillance systems. Sold as tools to reduce crime and manage traffic, these systems threaten civil liberties. They track movements, identify faces, and build the infrastructure for autocrats to silence dissent at scale.

This month’s counter-story is just as powerful. NVIDIA committed $26 billion to open AI models and partnered with Mistral, the French lab building some of the strongest open-weight models outside of China. Meanwhile, Microsoft open-sourced BitNet, a framework that lets 100-billion-parameter models run on ordinary CPUs. The direction is clear: powerful AI is moving onto personal devices, where it can work without an internet connection and without exposing data to anyone.

At SXSW, the world’s largest creative festival, HRF’s AI for Individual Rights Program brought this future to life. Our ESC TYRANNY booth showed what happens when activists from Togo, Venezuela, and Russia get their hands on AI: they vibe code resistance websites, build tools for movements under siege, and turn plain-language prompts into working infrastructure. Thousands of visitors stopped by our exhibit to try it themselves. Many had never written a line of code, but walked away having built something real.

We close with reflections from HRF’s Chief Strategy Officer Alex Gladstein on what happens when individuals control their own intelligence. OpenClaw, an open-source AI assistant, lets anyone execute complex tasks through simple prompts. Pair it with Bitcoin for censorship-resistant money and Nostr for censorship-resistant communications, and individuals gain capabilities that once required large teams and significant resources. The moment to jump in is now.

The Latest in AI for Repression

AI-Powered “Smart City” Risks Across Africa

A new report authored by researchers from the African Digital Rights Network, maps the spread of Chinese AI surveillance across eleven African countries. Algeria, Uganda, Nigeria, Rwanda, Zimbabwe, and six others have collectively spent more than $2 billion on “smart city” technologies: CCTV networks, license plate recognition, and AI-powered data analysis. The systems are marketed as tools to monitor roads and reduce crime, but the pattern is consistent: governments buy Chinese surveillance hardware under the banner of public safety, then point it at their own citizens. As these systems scale, they normalize mass monitoring and erode fundamental freedoms across the continent.

China Restricts Access to AI Agent OpenClaw

Chinese officials warned state-run enterprises and government bodies against installing OpenClaw on office computers without prior approval. The restriction extends to personal phones connected to workplace networks and even to the families of military personnel. The warnings come as OpenClaw sweeps rapidly across China. Schoolchildren, retirees, and professionals are all trying to use it to automate daily tasks and accelerate productivity. Reuters reported that the agent has “gone viral” across the country. OpenClaw does carry real security risks if deployed carelessly. But Beijing’s response reveals a familiar reflex: when an open, decentralized technology empowers individuals, the CCP moves to restrict access.

Iran Surveils Population with Russian Facial Recognition Tech

A joint investigation by Le Monde, Forbidden Stories, and other media partners revealed that Iran secretly acquired FindFace, a Russian-made facial recognition system developed by NtechLab, a company sanctioned by the EU in 2023 and blacklisted by the United States in 2024. Iranian front companies purchased the software in 2019 and distributed it to regime security services. FindFace matches faces across street cameras, identity databases, and social networks, giving the regime the ability to identify and track protesters with unprecedented precision. Russia has used the same software to hunt down supporters of murdered opposition leader Alexei Navalny and independent journalists.

China’s AI-Powered Operations to Target Dissent

OpenAI banned a ChatGPT account linked to Chinese law enforcement that was planning what it called “cyber special operations” that included Japanese Prime Minister Sanae Takaichi among its targets. The operations sought to generate fake content, impersonate officials, and organize coordinated harassment campaigns against critics inside China and abroad. When ChatGPT refused certain requests, the operator switched to locally deployed AI models to carry out the attacks faster and at greater scale. The case reminds us that authoritarian states are weaponizing AI not just for surveillance, but for active manipulation of public narratives across borders.

Guinea Expands Video Surveillance Network with Chinese Financing

Guinea's military junta, led by Mamady Doumbouya, approved a $56 million loan from China's Export-Import Bank to build a nationwide “safe city” surveillance program. The system will install 324 video surveillance sites across eight cities, with centralized command centers analyzing feeds in real time and police devices streaming live camera alerts. Guinea’s junta banned all protests in 2022, has used gunfire against demonstrators, and dissolved 40 political parties earlier this month. A regime that shoots protesters is now buying the tools to identify them by face.

Recommended Content

Human Rights at Risk in the Sprint Toward AI Sovereignty

Writing in Just Security, technology and democracy researchers warn that the global push toward national “AI sovereignty” carries serious human rights risks. When authoritarian states control their entire AI stack, they gain new tools for censorship, surveillance, internet isolation, and propaganda. Thailand, Vietnam, and China are already using sovereign AI infrastructure to expand mass surveillance and political control. While “sovereign AI” in the context of individuals controlling their intelligence is a noble goal, and one that HRF’s AI program is focused on, in the context of regimes developing their own AI, it can lead to a dramatic loss of civil liberties.

The Latest in AI for Freedom

NVIDIA and Mistral Advance Open AI for Global Access

NVIDIA committed $26 billion over the next five years to building open AI models, and the company partnered with Mistral AI, a French lab developing some of the most advanced open models outside of China, to co-develop them. Mistral also launched Mistral Small 4, an open model capable of general chat, coding, agentic workflows, and complex reasoning. This matters because open models run locally, privately, and offline, and they can be accessed through encrypted services like Maple without retaining conversation logs or user data. For human rights defenders working under surveillance, open models should be a security north star.

HRF Hosts ESC TYRANNY at South by Southwest

HRF brought AI for freedom to life at this year’s SXSW. Our ESC TYRANNY booth showed what "vibe coding" looks like in practice: activists in Togo building websites like Free-Togo to mobilize against the Gnassingbé regime, and movements from Uganda to Iran deploying Bitchat, a vibe-coded private messaging app that works offline, enabling people to communicate and share information during internet shutdowns. Visitors stepped up and built their own sites using nothing but plain-language prompts. Many walked away having seen, for the first time, how AI can turn the needs of human rights defenders into working infrastructure while saving significant time and costs.

Microsoft Launches BitNet to Advance Local AI’s Power

Microsoft Research open-sourced BitNet, a framework that reduces AI models to 1-bit precision, allowing 100-billion-parameter models to run on standard CPUs with six times the speed and 82% less energy than conventional approaches. No costly GPUs or internet connection required, and no data will leave the device. For dissidents working during internet shutdowns or under surveillance, BitNet is another step toward powerful AI that can run locally on a laptop, entirely offline, entirely private.

NVIDIA Developing an Open-Source AI Agent

NVIDIA introduced NemoClaw, an open-source enterprise AI agent. The move follows the lead of OpenClaw, which NVIDIA CEO Jensen Huang called “the most important software release probably ever.” NemoClaw will autonomously process emails, schedule tasks, analyze data, and automate workflows, with security and privacy tools built in. The key detail: It is open source. Developers can inspect the code, modify it, fork it, and deploy it for projects that have nothing to do with enterprise IT. In great news for civil liberties and individual sovereignty, open-source agents are becoming the norm, not the exception.

PayPerQ Launches an Encrypted AI

PayPerQ (PPQ) launched end-to-end encrypted AI models accessible without an account and payable with Bitcoin over Lightning. The service mixes user queries together, making individual conversations harder to trace. PPQ does not require registration, credit card, or identification. Users pay per query with satoshis (using any method that speaks Lightning: ecash, ark, and spark would all work) and get access to frontier models including GPT-4o, Claude, and DeepSeek. PPQ now also provides encrypted conversations with top open models like GPT-oss, Kimi, DeepSeek, and Llama. These models can plug into AI agents like OpenClaw, enabling private, automated workflows. For a human rights defender operating under surveillance, the difference between a chatbot that knows your name and one that does not could be the difference between safety and a prison cell.

Google Releases Dataset to Improve AI in Sub-Saharan Africa

Google Research released WAXAL, an open-source speech dataset of thousands of hours of recorded and transcribed speech in 27 African languages previously ignored in AI ecosystems. The dataset includes Wolof, Fon, Bambara, Igbo, and dozens of other languages spoken by over 100 million people. The dataset will enable developers to build voice assistants and transcription tools in languages that frontier AI models currently cannot understand. For activists and organizations working in local languages under dozens of authoritarian regimes in the area, this marks a major step forward.

Recommended Content

In a recent thread on X, HRF's Chief Strategy Officer Alex Gladstein argued that OpenClaw represents a Satoshi Nakamoto-like turning point: the moment individuals gained the ability to control their own intelligence. The software will become easier to install, more secure, and will integrate with a broader freedom technology stack, including Bitcoin for censorship-resistant money and Nostr for censorship-resistant communications. A single dissident with a laptop will be able to accomplish what once required a team of engineers, and can do it on their own terms. Read his full reflections here.