 How’s this for a prompt for ChatGPT or Claude: “Give me a list of people who’ve criticized the regime in the last several months.” Artificial intelligence helps small teams accomplish work that once required large organizations. Unfortunately, that now includes secret police units in repressive regimes around the world. According to the London-based, dissident-run Iran International news site, the AI company Anthropic discovered that Iranian paramilitary and security units had used its Claude AI model to monitor Iranians’ social media posts to identify opposition and diaspora accounts. One Iranian unit analyzed hundreds of thousands of social-media posts and identified 39 opposition accounts to track. Other actors used Claude to develop tools for identifying and profiling users, including a malicious Firefox extension disguised as a prayer-times utility. Iranian units also used Claude to develop or improve a centralized surveillance case-management system. AI was not merely collecting information. It was helping the state organize its surveillance bureaucracy and decide whom to watch. The stakes of Iran’s domestic surveillance are deadly. Medical networks, whistleblowers, and unofficial leaks estimate that thousands of Iranian dissidents and protesters, most of them young men and women, have been killed in the streets or executed just before the beginning of the U.S.-Iranian war earlier this year. To its credit, Anthropic has aggressively countered this misuse of its product. In the Iranian operations, it identified and unplugged 16 Claude accounts operated by two units associated with Iranian paramilitary and domestic security agencies. Iran is far from alone in its authoritarian abuse of AI. Axios reports that government-linked actors in China and Mali have also used Claude to automate surveillance. A consultant working for Malian security authorities reportedly created a system that gathers information from mobile operators and builds dossiers on individuals. In China, an intelligence office dedicated to spying on religious organizations once required large teams of analysts in many offices. AI consolidated it to a single office capable of producing thousands of investigations each month. Chinese authorities also used Claude to evaluate politically sensitive social-media posts and identify people for “control” – potentially including coercive questioning and closer monitoring of their movements and communications. Anthropic says it banned every account associated with the surveillance operations that it uncovered and strengthened its safeguards. But the company acknowledges the limits of any response. In the case of Mali, authorities simply transferred their surveillance platform to locally operated AI models. The danger is therefore larger than one company or chatbot. As Anthropic threat-intelligence chief Jacob Klein told Axios, AI is making state surveillance cheaper and more efficient. A lone official or contractor can increasingly perform work that once required whole teams of programmers and intelligence analysts. Authoritarian governments have always wanted to monitor dissidents, journalists, religious minorities, and political opponents. AI is abolishing the practical constraints that once limited human surveillance. George Orwell’s nightmare vision of totalitarian surveillance in 1984 always faced the practical constraint of recruiting enough watchers to monitor people around the clock. For every subversive comment or passed note, the watchers of 1984 would have had to sit through thousands of hours of people eating breakfast, brushing their teeth, and walking to work. But AI repression doesn’t get bored. It needs no lunch breaks or naps. And it is here. Comments are closed.
|
Categories
All
|
RSS Feed