Article · 2026-08-03 · 6 moments

Import AI 467: Self-sustaining AI viruses; pacing AI progress; confusion about AI and creativity

When do we build the moon arcology? ✦ AI generated

GUEST Jack Clark (Import AI)GUEST Dwarkesh Patel GUEST Jack Clark (Import AI), quoting OpenAI and Henry Yuen HOST Jack Clark

Claim◆

As AI systems get smarter, they will monetize the same amount of compute far more effectively, so the price of compute will rise dramatically before roboticization of the supply chain eventually brings it down.

The author relays Dwarkesh Patel's argument that smarter AI monetizes compute better, so prices will climb (an H100 could rent for $250k+ a year), though this is a temporary state before roboticized supply chains bring costs back down near raw inputs.

transcript

Dwarkesh Patel: As AI models become smarter, they'll better monetize the same amount of compute. If a true human-level software engineer that could run on an H100 equivalent, at current market rates for software engineers, that H100 should rent for over $250k a year. That's 15x today's spot prices. The reason AI is relatively cheap right now, at least in comparison to human labor, is partly that it can't do a lot of things that top humans can do. At some point that will no longer be the case. And so using GPUs to make short-form video slop will just get priced out.

Data◆

Self-sustaining, self-replicating AI computer viruses are no longer theoretical; open-weight LLMs running locally on compromised GPUs can autonomously detect vulnerabilities, devise tailored attacks, and replicate.

Researchers at Toronto, Vector Institute, Cambridge, and ServiceNow built a proof-of-concept worm that runs an open-weight LLM on stolen GPU compute to autonomously find and exploit vulnerabilities and self-replicate, with about a 37% overall attack success rate.

transcript

Jack Clark (Import AI): 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. The worm parasitically uses compromised machines to run open-weight large language models (LLMs) to sustain its reasoning, or extend its reach for further attacks. ... The worm uses stolen computing power from compromised GPU nodes to host LLMs for generative reasoning. It then uses this reasoning to detect vulnerabilities and devise tailored attacks against additional targets, furthering its spread. The proof-of-concept operates using only an open-weight LLM running on a single, local GPU, with no reliance on vendor APIs that could be monitored or revoked.

supports · 1

Prediction

The future internet will be like a complex ecology of attacker and defender AI agents carving out their own ecological niches, which may require humans to deploy their own defender agents as digital white blood cells.

The author argues the emerging worm research shows AI agents achieving operational resilience through decentralized swarms that resist any single point of control, framing the future internet as an ecology where humans may need autonomous defender agents.

transcript

Jack Clark (Import AI): 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. This may mean that humans need to create their own AI agents which they release onto the internet to serve as kinds of white blood cells against the adversary models. ... Despite the inherent fragility of individual exploitation attempts, the worm agent achieves operational resilience by continuously self-replicating into a swarm—a decentralized collective of independent agent replicas acting concurrently across the network. ... The worm operates in a fully decentralized manner, and no single point of control can be taken offline to interrupt its spread.

extends · 1

Claim

The world's leading AI companies ask the US government to support an international effort to deliberately pace the frontier of automated AI development, because capability growth may outrun our ability to control resulting systems.

Roughly 1337 signatories across leading Western labs (OpenAI, Anthropic, Google DeepMind, Thinking Machines, Meta, Safe Superintelligence Inc.) request US support for an international effort to develop technical and governance tools for pacing automated AI progress.

transcript

Statement signatories (OpenAI, Anthropic, Google DeepMind, Thinking Machines, Meta, SSI): AI could help create a dramatically better future, but that outcome is not guaranteed. The world's leading AI companies believe they could be close to automating AI research. It is hard to predict exactly how much this will accelerate AI progress, but there is a real risk that capability development rapidly accelerates beyond our ability to understand or control the resulting systems. ... We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.

supports · 2

Data

Today's AI systems are extraordinarily capable engineers but lack tasteful, original creativity — they fail to produce research of top ML conference caliber because they commit to narrow research paths early and cannot reverse out of unpromising approaches.

A shadow-evaluation study giving frontier agents unpublished NeurIPS 2026 research questions found both papers rejected for no novel contribution, poorly motivated experiments, and impenetrable prose, suggesting AI agents lack the creative insight needed to move the field forward.

transcript

Authors of the shadow-evaluation study (Princeton, Cornflower Labs, UK AI Security Institute, et al.): 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 [human] authors rejected both papers. The Personas paper was scored a 2 ("Reject"), and the TabPFN paper was scored a 1 ("Strong Reject"). Both reviews highlighted the same failures: poorly motivated data and experiments, no novel contribution, and impenetrable prose.

explains mechanism · 1supports · 1

Data

While AI systems struggle with original research creativity, OpenAI has solved ten verified open problems in mathematics and theoretical computer science using an internal version of Astra, signaling AI's arrival in domains where verifiability replaces subjective taste.

OpenAI's internal Astra model solved ten open problems across geometry, coding theory, complexity, group theory, cryptography, and combinatorics, with experts confirming their significance — read either as emerging creative intuition or as powerful verifiable engineering.

transcript

Jack Clark (Import AI), quoting OpenAI and Henry Yuen: These problems span high-dimensional geometry, coding theory, arithmetic circuit complexity, group theory, operator algebras, quantum complexity, lattice cryptography and extremal combinatorics. All of these problems are of substantial interest to their respective mathematical communities, and several are of broad interest across mathematics as a whole. ... New circuit lower bounds? A simple, easy-to-describe non-sofic group? Hardness of approximation for CVP without needing a unique games-like conjecture? I didn't just hear about these problems from my friends or from seminars. I feel their importance in my bones; I deeply care about the answers to these questions.

Highlight slides

Autonomous AI worms are no longer theoretical✦ from: Self-sustaining, self-replicating AI computer viruses are no longer theoretical; open-weight LLMs running locally on compromised GPUs can autonomously detect vulnerabilities, devise tailored attacks, and replicate.Smarter AI Drives Up Compute Prices✦ from: As AI systems get smarter, they will monetize the same amount of compute far more effectively, so the price of compute will rise dramatically before roboticization of the supply chain eventually brings it down.Proof-of-concept achieves ~37% attack success rate✦ from: Self-sustaining, self-replicating AI computer viruses are no longer theoretical; open-weight LLMs running locally on compromised GPUs can autonomously detect vulnerabilities, devise tailored attacks, and replicate.The Price Peak Is Temporary✦ from: As AI systems get smarter, they will monetize the same amount of compute far more effectively, so the price of compute will rise dramatically before roboticization of the supply chain eventually brings it down.

Related episodes

Audio · All-In Podcast

OpenAI Misses Targets, Codex vs Claude, Elon vs Sam Trial, Big Hyperscaler Beats, Peptide Craze

→

Audio · Lex Fridman

#465 – Robert Rodriguez: Sin City, Desperado, El Mariachi, Alita, and Filmmaking

→

Video · Machine Learning Street Talk

Intelligence is collective, not artificial — Prof. Michael I. Jordan (UC Berkeley / Inria)

→