A joint study by the University of Edinburgh, the University of Cambridge, and the University of Strathclyde reviewed 97,895 discussions on underground forums from 2022 through late 2025 and reached a conclusion that runs against common fears in cybersecurity: AI has not meaningfully lowered the technical barrier for hackers, nor has it caused a major shift in established criminal business models or operating practices.
The research, led by Edinburgh security researcher Ben Collier, focused on a question that has hung over the sector since ChatGPT entered public use: how much stronger, in practical terms, have hackers become with generative AI. The answer from the forum data was restrained. The biggest visible changes were not in elite intrusion work, but in areas that were already highly automated.
Forum status systems are colliding with AI-written spam
The study argues that underground forums function as more than marketplaces for stolen data. They also operate as a reputation economy, where trust, status, and influence are tied to visible technical skill. Users build standing through posts, point systems, contests, and peer recognition, in a way that the report compares to communities such as Stack Overflow or GitHub.
That helps explain why AI-generated filler content has drawn hostility. Forum users quoted in the report complained that some people were using AI to write posts instead of contributing directly, and others said they came to these spaces to interact with real people, not chatbots. Collier said the tension is structural: AI weakens one of the signals members use to present themselves as highly skilled operators.
Measured impact appears in scams, bots, and bulk content
According to the study, AI’s clearest effect was concentrated in lower-skill, volume-driven activity, including SEO fraud, social media bots, and some forms of romance scams. In those areas, AI can help produce text at scale, smooth language, or make social engineering messages more convincing. That is different from the upper end of offensive capability.
In fields that define technical depth, such as penetration testing methods, vulnerability research logic, and 0-day work, the researchers said they found little measurable evidence that AI had changed outcomes in a major way. The report pushes back on the idea that generative models have broadly opened the door for inexperienced users to perform advanced hacking.
Underground forums face the same pressure seen across the web
The report also describes a problem that extends beyond cybercrime communities. AI slop reduces discussion quality, drowns out credible signals, and can drive away human participants. It notes that some forums are also seeing fewer visitors as Google AI search summaries pull attention away from source communities.
Even inside hacker circles, attitudes toward AI were not uniform. Some users accepted tools for grammar fixes or structural editing, while rejecting fully AI-written posts. The divide was less about total rejection and more about where the line is drawn between assistance and replacement.