HRF's AI for Individual Rights Newsletter #16

• THE LATEST IN AI FOR REPRESSION
• RECOMMENDED CONTENT
• THE LATEST IN AI FOR FREEDOM
• RECOMMENDED CONTENT

Welcome to the 16th edition of HRF’s AI for Individual Rights newsletter.

Chinese tech giant Huawei is bidding to expand Egypt’s AI capacity, bringing more powerful artificial intelligence (AI) tools into the hands of a dictatorship. In a country where officials already closely monitor and suppress dissent, that expansion could make surveillance faster and more sophisticated.

Elsewhere, technology is expanding human potential. US software company Perplexity launched Portable Computer, an AI agent designed specifically to operate privately on a user’s device by default (local-first). This points to a future (more broadly being accepted by major corporations) where anyone could have an AI agent that works for them, that operates away from any company or regime that tries to watch them.

We close with a tutorial from Freedom Tech educator Ben Perrin, also known as Sovereign Sessions and BTC Sessions, who shows users how to securely access an AI model running locally (and therefore privately) from different devices.

§THE LATEST IN AI FOR REPRESSION

§Egypt May Work with Chinese Company Huawei to Expand AI

Chinese tech company Huawei is bidding to build out Egypt’s AI infrastructure for surveillance, military, and other public sector uses. The proposal would include providing Egypt with more than 2,000 advanced AI chips that could help the country build the computing power to run high-performance AI systems. The deal could also involve iFlytek, a Chinese company that sells vehicle and facial recognition software and which the US sanctioned for facilitating Beijing’s mass surveillance of Uyghurs. If the deal moves forward, Egypt’s fully authoritarian regime will have a greater technical capability to surveil its population and any dissidents within it.

In context: Just days after the proposal, Xi Jinping visited Egypt to discuss closer bilateral ties, potentially including future AI-related collaborations. Together, this visit and Huawei’s bid show how seriously China is pushing for global AI dominance. If successful, deals like these could give China’s authoritarian regime a foothold in the AI systems and standards adopted by other repressive regimes.

§Indonesia Expands AI to Surveil Its Borders

Indonesia is expanding the use of AI and drones to patrol its borders and monitor foreign nationals. To do so, the country’s immigration agency will work with a local technology institute to develop a “Digital Fence” system, in which drones will watch unofficial crossing routes and other hard-to-reach areas. Although the system is being built for immigration enforcement, it also expands the hybrid authoritarian regime’s broader surveillance capacity. There is also no guarantee it will remain limited to monitoring borders.

In context: Indonesia violently cracked down on nationwide protests just last year, with at least 10 people killed and more than 3,000 detained. As the state builds more powerful surveillance tools, they could double as mechanisms of repression to use against future dissent. 

§Russian AI Influence Campaign Using ChatGPT

US tech company OpenAI discovered and banned a cluster of ChatGPT accounts linked to a Russian influence operation. The accounts were using ChatGPT to generate social media comments across X, LinkedIn, Facebook, Substack, and Telegram. The posts promoted the International Burke Institute, a suspected Russian front posing as an independent think tank. This organization created an index that allegedly praises Russia and negatively portrays Western countries, while posting copied and misattributed materials attacking governments critical of Russia. The AI-assisted social media campaign did not reach a large audience, but it is a reminder that influence actors in authoritarian regimes can use AI to cheaply and quickly scale propaganda. 

§China’s AI Olympics and Lead in Humanoid Robots 

China held its second World Humanoid Robot Games in Beijing, showcasing its dominance in humanoid robotics. In the games, more than 2,000 robots competed in races, martial arts, obstacle courses, weightlifting, and more. And the progress in engineering was “staggering,” according to international experts. The fastest robot, Tiangong Ultra, ran 100 meters in 8.64 seconds — 0.94 seconds under Usain Bolt’s 9.58-second human record. The best standing jump rose from 0.96 meters to 3.4 meters. The games were a fun showcase, but they were also a progress report. The pace of improvement suggests that China’s humanoid capability is compounding faster than expected.

Why this matters: For an authoritarian regime, progress like this can have unsettling implications. China is already using AI-powered robot police with facial-recognition cameras to patrol public spaces. As humanoids add mobility and dexterity, the gap between a “robot that plays games” and a “robot that patrols, follows, and records” may narrow quickly, to the detriment of dissent.

§Nigerian Engineers Urge the Regime to Use AI for Predictive Policing and Surveillance

A local branch of the Nigerian Society of Engineers, a professional engineering association, is calling for Nigerian officials to adopt AI for predictive policing and surveillance. The branch chairman argued that AI-powered pattern analysis, smart cameras, drone surveillance, and systems that predict crime before it happens could help address the country’s insecurity, including kidnapping, banditry, Islamist insurgency, and communal violence. Those threats create an understandable case for stronger security tools. But once built, such tools could quickly be turned against critics and protesters. 

In context: Nigeria’s hybrid authoritarian regime has systemically repressed dissent, intimidated critics, and violently suppressed protests. Developments in AI-powered surveillance and predictive policing could give it powerful tools to stop dissent even before it evolves. 

§China’s Censorship in Tibet Amid Nepal–Tibet Floods 

China is reportedly censoring and suppressing information about the recent devastating floods affecting Tibet, a territory under Chinese Communist Party (CCP) occupation. Videos posted by locals that show the disaster have been repeatedly removed from Chinese social media platforms. Journalists have reportedly been barred from the disaster zone and blocked from interviewing families. At least 10 people have been punished for alleged “rumormongering.” Some locals are afraid to post their testimonies online, leaving the public to rely on CCP-controlled media. In such a dire time, China’s censorship risks blocking rescue efforts and leaving people without the information to find missing loved ones.

In context: China uses AI to automatically censor its social media platforms, often to ensure posts reflect a positive narrative about the regime. This incident demonstrates how authoritarian digital censorship, especially if it is scaled with AI, is a violation of freedom of expression that can also endanger lives.

§RECOMMENDED CONTENT

§The Kremlin Is Cultivating an AI-Trained Global Media Cadre for Lasting Influence

The Institute for the Study of War, a think tank based in Washington, D.C., released an article exposing how the Kremlin trains international journalists to use AI to generate Kremlin-aligned content. State-run training programs across Africa and Asia teach participants to discredit the West, validate the Kremlin’s full-scale invasion of Ukraine, and obscure the regime’s weaknesses. With AI, this volume and reach of the content could multiply. Read the article here.

§THE LATEST IN AI FOR FREEDOM

§Perplexity Launches Local-First Agent Stack

US software company Perplexity launched Portable Computer, an AI agent designed to operate locally by default (local-first, and therefore privately when kept on-device). Currently, users need to set up the agent on an NVIDIA DGX Spark, a high-end personal computer that costs more than $4,000. The agent is then powered by an AI model that runs directly on the device. Perplexity says it can run without a third-party server or internet connection. If a task requires a more powerful model or access to the web, the agent will ask permission before sending any information off the device. While its setup is expensive, Portable Computer is a major step toward powerful AI agents that work only for the individual. Hopefully, it’s just the beginning of easy-to-use agent tools where sensitive files, prompts, and day-to-day computing remain under the user’s control, away from any company or government. 

Why this matters: AI agents are incredibly useful to human rights defenders. And when they run privately and completely under their control, they become tools for genuine individual autonomy. The barriers to entry to Portable Computer are still high, but this development shows that companies are building in the right direction.

§Z.ai Releases Open-Weight AI Model GLM-5.3 Flash

Chinese AI technology company Z.ai has released GLM-5.3 Flash, a new open-weight model. Open-weight models are ones that users with sufficient hardware can download, modify, and run on their own devices. GLM-5.3 Flash has received praise for delivering strong performance in coding and agentic tasks while using less computing power. When users ask GLM-5.3 Flash a question, it uses only a small fraction of the model at a time. This design makes the model cheaper to run and capable of running locally on devices such as an NVIDIA DGX Spark. Users without a powerful enough device can interact with the model privately through Maple AI, a service that lets people access open-weight models in an end-to-end encrypted manner. GLM-5.3 Flash gives more individuals and organizations a capable, open-weight model they can use privately.

Important: Since this model was developed in China, it may reproduce the CCP’s censorship. Be wary when relying on its outputs, especially for politically sensitive information.

§Apple Releases New Mac Studio and Mac Mini for Local AI

Apple has released a new generation of devices specifically designed to run powerful open-weight AI models locally. Leading the lineup are the Mac Studio M5 Max and M5 Ultra, which offer enough memory (capacity) to run some of the world’s most capable open-weight models entirely on-device. They are expensive, ranging from $2,499 to over $18,000, but Apple says they represent a “monumental leap in AI performance” over the previous generation. A human rights organization could use them to run extremely powerful AI, such as GLM-5.3 Flash, privately, on hardware that never leaves the office.

Apple also released cheaper devices, the Mac minis M6 and M5 Pro, starting at $899, which can run smaller AI models at a significantly faster speed than previous versions. While these still may be out of the budget for individual dissidents, they show how quickly local AI hardware is becoming more capable and accessible. It’s exciting to see one of the world’s largest tech companies building computers that make AI more private and user controlled. 

§OpenClaw Releases Its Largest Update Yet

OpenClaw, the open-source agent that shocked the AI world with its capabilities at its launch, has released OpenClaw 2.0. Developers say they simplified the setup and improved long-term memory, security, plugins, team collaboration, and more. Some users have criticized the release as underwhelming, but these improvements may still give activists an easier-to-use, capable agent that can work for them.

Why this matters: Because OpenClaw is open-source, users can use it with any AI model, including privacy-protecting and local models. They can also self-host the agent on their own devices. That gives users more control over how the agent works and where their data goes. Even if OpenClaw 2.0 falls short of expectations, it’s encouraging to see people working to improve tools that can empower the individual. 

§Tailcat Makes Private AI Easier to Access Anywhere

Tailscale, a company that builds private networking tools such as VPNs, released Tailcat, an open-source tool that lets two computers make a temporary, encrypted connection without a complicated setup. Why would anyone want to do this? Imagine you have an AI agent running locally on your computer. Typically, you would need to be on that computer to interact with the agent, but with tools like Tailcat, you can communicate with the agent from other devices. The company says the connection is encrypted, so the data traveling between devices is protected from outsiders. Tailcat is designed for quick, temporary connections, with no account needed, for short-term needs.

Why this matters: Tailcat could make it easier to run private AI models or agents that are accessible across multiple devices and locations.

§RECOMMENDED CONTENT

§Your Own Private AI Server: Access LM Studio Anywhere

In this tutorial, Freedom Tech educator Ben Perrin (also known as Sovereign Sessions and BTC Sessions) teaches users how to take private AI with them anywhere. He walks users through turning a computer running LM Studio — an application for running AI models locally — into a private server. Then, he explains how to use Tailscale, a VPN that privately links devices, to securely access that AI from another device. The result is private AI that is accessible anywhere. Watch the tutorial here.