# Import AI 475 covers swarm scaling, DeepMind biology watermarks, and AI science economy

Digest AI · Research · published 2026-10-05T12:32:30Z

Canonical: https://digestai.news/story/import-ai-475-covers-swarm-scaling-deepmind-biology-watermarks-and-ai

## Summary

Jack Clark's Import AI newsletter issue 475 surveys three research areas. Toby Ord analyzes swarm scaling, finding that four-agent swarms achieve similar performance to single agents using half the tokens per agent but twice the total tokens, yielding speed gains through parallelization. A "stepping on toes" parameter shows diminishing returns: 10x more agents deliver only 3x-5x performance gains. Ord suggests this makes recursive self-improvement more likely.

Google DeepMind introduces SynthID Bio, a family of watermarking methods for synthetic biology. In wet-lab tests on three target proteins (VEGF-A, SARS-CoV-2 spike protein RBD, PD-L1), watermarked designs matched unwatermarked versions on hit rate, binding affinity, and sequence diversity. The approach aims to strengthen biosecurity and scientific integrity.

C5R Corp's SciUniverse benchmark tests AI systems on 92 tasks across 17 families in automated labs. Claude Fable 5.1 (xhigh) leads at 45.3% pass rate and $40.61 cost per task, followed by GPT-5 Astra (xhigh) at 32.5% and $52.37, then Claude Opus 5 (xhigh) at 30.5% and $46.31. DeepMind researchers separately propose an "Automated Scientific Economy" with four components: proof of ideation, ex-ante evaluation, brokerage and trade, and validation payout, to match AI-generated ideas with scarce physical resources.

CSAIP polling of 2,498 Americans shows 61% say voluntary AI company commitments are "not enough," including 53% of Trump voters. Fifty-four percent want government to set and enforce AI rules (61% Harris voters, 48% Trump voters).

## Key points

- Four-agent swarms use half the tokens per agent but twice total tokens for same performance, enabling speed via parallelization
- DeepMind's SynthID Bio watermarks synthetic biology designs; wet-lab tests on three proteins show matched performance
- SciUniverse benchmark: Claude Fable 5.1 leads at 45.3% pass rate, $40.61 per task across 92 automated lab tasks

## Why it matters

Swarm scaling suggests faster AI capability gains; biology watermarks address biosecurity; automated lab benchmarks track AI's growing scientific capabilities; public polling shows demand for AI regulation exceeds current policy.

## Sources

1. [Import AI 475: Swarm scaling; Google DeepMind watermarks biology; and the AI science economy](https://jack-clark.net/2026/10/05/import-ai-475-swarm-scaling-google-deepmind-watermarks-biology-and-the-ai-science-economy) (Import AI, 2026-10-05)

Part of the developing story: [DeepMind Bio Watermarks and AI Science](https://digestai.news/thread/synthid-bio-watermarking-introduced-for-ai-generated-biology) (2 stories)

## Cite

Digest AI, "Import AI 475 covers swarm scaling, DeepMind biology watermarks, and AI science economy", 5 October 2026, https://digestai.news/story/import-ai-475-covers-swarm-scaling-deepmind-biology-watermarks-and-ai

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/import-ai-475-covers-swarm-scaling-deepmind-biology-watermarks-and-ai.json