Cryptographer and Malware Researcher Clash Over AI's Impact on Cybercrime

On October 30, noted cryptographer and Johns Hopkins professor Matthew Green and MalwareTech, a security researcher who has long tracked botnets and worms, engaged in a heated multi-round debate on X over whether AI will reshape the cybercrime landscape, with the discussion spilling into a side thread about the long-term demand for AI vulnerability research.

Confirmed

• Green's core concern is that open-weight models with safety guardrails stripped out via "abliteration" (such as abliterated GLM and Qwen) could enable people without specialized skills to carry out cyberattacks, significantly lowering the barrier to entry.
• MalwareTech countered that it is hasty for academic researchers to draw conclusions about "advanced AI capabilities changing the threat landscape" right after downloading guardrail-removed models. He has tracked botnet and worm threat actors full-time since 2015, and offered IoT malware as an analogy—it has become background noise, like spiders in the house, not worth the effort to eradicate—implying that deploying small open-source models on IoT/edge devices offers limited value. He also denied "downplaying AI."
• In another post, MalwareTech argued that AI vulnerability research is in a novelty-driven "honeymoon period," with money pouring into frontier models even from people who previously paid no attention. But long-term demand is limited: outside elite researchers, vulnerability research only leads to two paths—VR work for security companies or bug bounties—and demand could fall off a cliff once the hype fades.
• Researcher davidmanheim pushed back on that judgment: as attack costs drop, software is unlikely to return to an era where "attacks were hard and most targets were safe by virtue of being unimportant"; enterprises' demand for continuous security scanning may be more resilient than expected.
• Green clarified his own position: he is glad to see the security community pushing AI labs to correct past bad decisions, agrees that security and sandboxing matter, and does not consider himself a doomer. But he objects to bundling that stance with an attitude that "new model capabilities are no big deal," and believes the security community's criticism of AI labs carries a streak of insecurity.

Why it matters

The debate crystallizes two typical positions within the security community on AI: one side stresses the real risks posed by abliterated open-source models and rising capabilities, while the other, drawing on years of frontline threat tracking, believes the threat narrative is overblown. This divide directly shapes how the industry judges AI security investment and the trajectory of the vulnerability research market, and Green's reflection that "criticism carries insecurity" offers the community a more introspective angle for discussion.

2026-10-30 ~ 2026-10-30 · 9 related posts

• [source] MalwareTech: AI vulnerability research demand is a honeymoon bubble that will collapse — MalwareTechBlog · 2026-10-30
• Security researcher and cryptographer clash over whether abliterated GLM experience predicts AI threats — matthew_d_green · 2026-10-30
• [source] Cryptographer Matthew Green and malware researcher MalwareTech clash over whether AI will reshape cybercrime — matthew_d_green · 2026-10-30
• Counterpoint: if attacks keep getting cheaper, endless scanning demand may be durable — davidmanheim · 2026-10-30
• MalwareTech: IoT malware is like spiders in your house, not worth clearing out — MalwareTechBlog · 2026-10-30
• Cryptographer Matthew Green: backing AI safety critics, but model capabilities are no small deal — matthew_d_green · 2026-10-30
• Cryptographer Matthew Green slams Anthropic's 'model escape' framing of agent worm research — matthew_d_green · 2026-10-30

2 near-duplicate retellings: matthew_d_green · matthew_d_green