Arc Institute

Overview

The Arc Institute is a nonprofit research organisation based in Palo Alto, California, focused on long-timescale biomedical research combining wet-lab biology with AI foundation models. Arc is the primary institutional home of the Evo genome-scale language-model series (Evo 1, Evo 2) — foundation models trained on DNA / genomic sequences that can generate novel biological constructs. Arc frequently collaborates with Stanford and UCSF and publishes in Nature, Science, and Cell.

Timeline

• 2026-08-08-AI-Digest — Stanford researchers and Arc Institute publish a Science paper (Aug 6) documenting the use of Evo 1 / Evo 2 to design 16 novel bacteriophages that successfully killed E. coli in laboratory testing, at a ~5.6% hit rate against the candidate pool. First documented laboratory demonstration of foundation-model-designed viable viruses of any kind. Bacteriophages target bacteria not humans — paper scope is antibacterial-therapy design — but the technical shape (foundation-model design of viable genome-level constructs at nontrivial hit rate) is what the biosecurity discussion has been extrapolating from for two years. Reads with Anthropic‘s Claude Fable 5 biology-safeguard recalibration and OpenAI’s Astra Critical cyber pause the same week as a co-temporal biosecurity cluster.

Key Developments

1. Evo 1 / Evo 2 → AI-Designed Bacteriophages (August 6, 2026): Stanford + Arc Science paper demonstrates that foundation-scale genome language models can design viable novel viruses at laboratory scale. 16 successful bacteriophage designs killed E. coli out of the candidate pool (~5.6% hit rate). Load-bearing datum for the biosecurity discussion: the first primary-source lab demonstration replacing projection with replicated result. Anchors the MOC - Agent Security biosecurity thread against a specific published capability rather than a capability rumour.

Related

• MOC - Agent Security