Welcome to the 17th edition of HRF’s AI for Individual Rights newsletter.
In Taiwan, officials revealed that Chinese tech companies are ramping up efforts to recruit Taiwanese engineers and gain access to the island’s AI chip expertise. Officials investigated 166 companies operating with concealed links to China and 67 cases involving trade secrets, including the unauthorized transfer of critical technologies. Access to Taiwan’s chipmaking expertise could help China’s authoritarian regime strengthen the AI systems it uses for surveillance and control.
But technology flows can work for freedom as well. NVIDIA, the world’s leading AI chipmaker, is planning to acquire Hugging Face, the top platform for downloading and sharing open-weight AI models. Open-weight models are simply files that individuals can download, run, and then interact with on their own hardware, without sending their data to a third-party provider. Together, NVIDIA and Hugging Face plan to make open models easier for anyone around the world to access and use.
We include a new HRF-sponsored report from the Australian Strategic Policy Institute, “Warning signals: Venezuela and the risk of Chinese AI-enabled digital authoritarianism.” This report examines how well-positioned Venezuela is to adopt AI-powered surveillance and censorship tools from China and use them to repress its population.
The Latest in AI for Repression
China’s Undercover Chip Labs in Taiwan
Over the past six years, Taiwanese officials have reportedly opened more than 200 investigations into Chinese efforts to gain access to the island’s technology and AI chip expertise. Taiwan has investigated 166 Chinese-linked technology companies for illegally operating in the country, often using shell companies or foreign entities to conceal themselves. These companies then recruited Taiwanese engineers to gain access to their AI talent. Separately, 67 cases involve trade secrets and alleged theft or transfer of protected technology from Taiwan to China. Together, these efforts could help China’s authoritarian regime gain the know-how to strengthen its own chip industry and overall AI capacity.
Why this matters: In China, the regime uses AI to censor political dissent online, power millions of facial recognition cameras, and build tools that could flag political risk. But China currently lags behind Taiwan’s lead in chipmaking. By stealing this expertise, Beijing could more easily create its own advanced AI chips that supercharge its repressive efforts.
AI-Generated Campaign Promotes Female Genital Mutilation in Mali
In Mali, an anonymous social media campaign is using AI to generate graphics that promote female genital mutilation (FGM), the partial or total removal of female sexual organs. The posters contain headshots of religious and cultural leaders, who likely did not consent to be included, showing support for FGM. Graphics read “Yes to excision” (excision refers to FGM) beside images of razor blades and green check marks. For victims, this campaign is very disturbing. One woman who underwent FGM said, “We were all astonished, angry, and surprised to see men launching a campaign saying ‘Yes to excision.’ We thought it was becoming less of a taboo.”
In context: Mali’s official records say that 89% of women between 15 and 49 have been subjected to FGM. The United Nations recognizes this practice as “an act of violence against women and girls that impairs and undermines the enjoyment of their human rights.” It is disheartening and disturbing to see AI-generated content promoting a practice that harms millions of women.
Malaysia Eyes Huawei Chips for Sovereign AI Plans
Malaysia is reportedly considering buying AI chips from Chinese technology giant Huawei to power a $494 million AI initiative. According to people familiar with the matter, this deal is “intended to give the country greater control over national data.” But giving Malaysia’s hybrid authoritarian regime control over that data could also give it more power over its citizens. Paired with greater AI capacity from these chips, officials could more easily search databases, connect information, identify dissidents and their relationships, and quickly profile political opposition.
In context: Huawei is increasingly competing to supply AI chips globally, as seen in its recent bid in Egypt. If Malaysia moves forward with a deal with Huawei, Bloomberg says it would mark the first known instance of a foreign country opting for Chinese AI chips over American ones. That would strengthen the position of Huawei, a company linked to the Chinese Communist Party, in the race to supply AI infrastructure worldwide.
Serbian Activists Targeted with Spyware
Digital rights group SHARE Foundation has documented at least 14 people in Serbia who have been targeted with advanced spyware capable of exposing everything on their devices since the beginning of 2026. The targets include student protesters, activists, an opposition member of parliament, and a local councilor. In one case, a pro-democracy student’s iPhone was remotely infected with Pegasus spyware, a tool that gives outsiders complete access to the device without the victim even clicking a link. Evidence suggests that Serbian state officials were behind at least some of the attacks, and many occurred around Serbia’s local elections in March. This could be a sign that officials were trying to maintain political control by keeping tabs on the people most likely to challenge it.
In context: AI could accelerate both the development of spyware like Pegasus and the expansion of its surveillance powers. It is concerning to think how hybrid and fully authoritarian regimes equipped with AI could deploy spyware to track critics and their networks at unprecedented scales.
Recommended Content
Warning signals: Venezuela and the risk of Chinese AI-enabled digital authoritarianism
The Australian Strategic Policy Institute, an independent think tank, released a report with support from HRF detailing how Venezuela might integrate Chinese AI into its existing surveillance system. The dictatorship already uses Chinese-built cameras, telecom infrastructure, and command centers. Additionally, it has already signed a deal with a Chinese tech firm to integrate AI that could make these intelligence systems more repressive. Read the report here to learn more about how Chinese AI could deepen digital authoritarianism in Venezuela.
The Latest in AI for Freedom
NVIDIA Acquires Hugging Face to Advance Open-Weight AI
US tech giant NVIDIA announced plans for a $13 billion acquisition of Hugging Face, the leading platform for downloading open-weight models. Open-weight models are ones that users can download onto their hardware and run themselves, privately. NVIDIA claims that Hugging Face will remain an open ecosystem for downloading models, while the acquisition will give the platform more engineering support, infrastructure, and global reach. NVIDIA CEO Jensen Huang explains in the announcement, “Together, we will make AI more open, more capable and more accessible to people and institutions around the world.”
Why this matters: Open-weight AI can greatly empower individuals. Instead of depending on an AI company that can suspend accounts or collect user data, people can run open-weight models locally on their own devices, keeping sensitive files and conversations private, without an internet connection or risk of a shutdown. This collaboration between two of the leading forces in open-weight AI will, hopefully, further this freedom.
Open-Source AI Summit in San Francisco
Presidio Bitcoin, a hub for builders working toward a more private and decentralized future, is hosting the Open Source AI Summit in San Francisco on Sept. 10 and 11. Researchers, developers, and entrepreneurs from leading AI organizations (including Hugging Face, Google, Tinfoil, Maple, and Spiral) have gathered at the Presidio. Together, they are exploring how AI can remain open and developed to protect privacy, why individuals have a right to access AI, and the philosophical debates over closed versus open AI systems. HRF’s own AI technical lead, Justin Moon, is among the speakers, explaining how open and privacy-protecting AI is essential for human rights defenders challenging authoritarian regimes.
Why this matters: Events like this foster collaborations that ensure the future of AI is not controlled by a few companies or governments, but remains open and private, ultimately serving the individual.
Open Technology Offers Relief during OpenAI, Anthropic, and xAI Outage
Last week, AI models from US tech companies OpenAI, Anthropic, and xAI all experienced outages within hours of each other for reasons not fully clear. But while people lost access to their closed-sourced models for hours in some cases, open technology provided an alternative. Anyone with an open-weight model downloaded locally could keep using AI. Another freedom tech tool, Routstr, also stayed up and running. Routstr lets people access open-weight models on Nostr, a decentralized protocol that uses a network of different relays (similar to servers). Because it doesn’t depend on a single company or server, there is no single switch that can simply shut the platform down.
Why this matters: Centralized AI creates a single point of failure. Users can lose access when a company’s servers go down, when it goes offline, or when it decides to restrict access. Open and decentralized alternatives make AI more resilient and give individuals more control over how they access intelligence.
Tether Releases Open Models for African Languages
Tether AI Research, an initiative from stablecoin issuer Tether, launched a family of open AI models designed to help with translations of Sub-Saharan African languages. The release included QVAC TranslatePsy-AfriSLM, which supports 19 African languages (such as Hausa, Amharic, Yoruba, and Lingala), and QVAC TranslatePsy-AfriNano, which supports 8 African languages. Crucially, these models are small enough to run locally on phones and laptops, making private AI translations far more accessible to human rights defenders and marginalized populations. Find them on Hugging Face here.
Why this matters: The needs of the hundreds of millions of people who speak these languages across Africa are often underserved by mainstream AI. These new models can help close that gap, while letting people translate sensitive information privately.
Vibe-Coded Website for South Sudanese Movement Sout Al Shaab
South Sudanese human rights activist and HRF Freedom Fellow Bol Deng Bol vibe coded (used AI to create) a site for his movement, Sout Al Shaab. The youth-led movement works to give South Sudanese people nonviolent ways to participate in public life, organize around political and social issues, and demand greater accountability from a repressive regime. Instead of needing to hire a developer or learn to code from scratch, Bol used AI to turn his ideas into a working website and give the movement a stronger online presence. Check it out here.
In context: In South Sudan, Salva Kiir’s regime restricts civic space through the absence of national elections, censorship, arbitrary detention, harassment, and violent crackdowns on peaceful protests. Sout Al Shaab is creating a peaceful way for citizens to speak out.
Perplexity Launches Hybrid Compute to Mix Local and Cloud AI Agent Use
US software company Perplexity has launched Hybrid Compute on Mac, a new system that lets Perplexity’s AI agent split tasks between powerful cloud models and open-weight models running locally on a Mac. Users can install an app on a Mac, download a local model, and then ask the agent to complete a task. Before information leaves the computer, a local privacy layer checks for sensitive data such as names, payment information, or passwords. This information is then masked or kept entirely on the device. Meanwhile, less sensitive parts of the task are handled by the frontier model running in the cloud. This way, users get the best of both worlds. They keep sensitive information on their computer while benefiting from the most powerful AI models in the cloud.
Important: Hybrid Compute is not fully open source, so its privacy protections cannot be verified by inspecting the entire system’s code. Use with caution.
Recommended Content
Owning Your Inference with Austin Kelsay
In a recent episode of the Tangents podcast, host Rod Roudi interviews Finite Co-Founder and Minister of Intelligence Austin Kelsay to explore the local AI ecosystem. They discuss how to get started running AI on your own hardware, how to best take advantage of privacy-protecting AI, and when it still makes sense to use frontier (but not private) models like ChatGPT. Watch it here.
