Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativity
Import AI · Jack Clark · 2026-08-03
Jack Clark's Import AI newsletter covers a functional AI-powered self-replicating worm prototype achieving ~37% end-to-end attack success, a statement from ~1,337 AI researchers requesting government mechanisms to pace frontier AI development, research showing current AI lacks creative research ability, and OpenAI's AI solving ten open mathematics problems.
Appears in
Extraction
Topics: ai-securityai-governanceai-research-autonomycompute-economicsai-creativity
Claims
- Researchers from the University of Toronto, Vector Institute, Cambridge, and ServiceNow have built a prototype AI worm that uses open-weight LLMs to detect vulnerabilities, exploit them, and self-replicate across GPU infrastructure with roughly 37% overall attack success rate.
- Dwarkesh Patel argues that compute prices could rise 10-15x as AI systems become capable enough to substitute for high-value human labor like senior software engineers.
- Approximately 1,337 senior AI researchers from major Western labs including OpenAI, Anthropic, Google DeepMind, Meta, and Safe Superintelligence signed a statement requesting US government support for international mechanisms to deliberately pace frontier AI development.
- Current frontier AI systems can perform sophisticated engineering tasks but fail to produce creative, original AI research at the level of top ML conference submissions, potentially slowing recursive self-improvement timelines.
- OpenAI used an internal version of its next major model to solve ten open problems spanning high-dimensional geometry, coding theory, group theory, quantum complexity, and other advanced mathematical fields.
Key quotes
We must prepare for autonomous generative adversaries. Artificial intelligence (AI) agents enable a fundamentally new threat: a worm that generates tailored attack strategies to each target it encounters.
While agents could solve the engineering problems necessary to do the research, they failed to produce original research at the caliber of a top ML conference.
The future internet is going to be more like a complex ecology full of attacker and defender AI agents than anything else; research like this shows how certain AI agents might end up carving out their own ecological niches, living off of infrastructure and self-replicating autonomously, beyond human control.