Hackers debate AI: curiosity, skepticism, and bespoke criminal models

A new study analyzing over 160 conversations across cybercrime forums over seven months offers a rare window into how criminal operators actually perceive AI. Researchers found the community split between genuine curiosity about offensive applications and persistent doubts about whether current tools deliver enough capability to justify the operational security tradeoffs they introduce.

Two parallel tracks emerged. Some actors focus on jailbreaking and abusing mainstream legitimate models, while others invest in building bespoke criminal-purpose models stripped of safety guardrails. Both approaches face skepticism from veteran forum members who question reliability, output quality, and whether AI meaningfully improves on existing tooling for core tasks like phishing, malware development, and social engineering.

The significance is in the framing: rather than treating AI-enabled cybercrime as an inevitable wave, the diffusion-of-innovation lens shows adoption is uneven, contested, and shaped by the same cost-benefit calculus any underground market applies to new tradecraft. For defenders and policymakers, that means the near-term threat landscape depends less on AI’s theoretical capabilities and more on which specific workflows criminals decide are worth the switch.