UK AISI: coding agents rebuilt simple biological AI tools from a paper in under 4 hours; data filtering safeguard undone for <$80
The UK AISI Chem-Bio team built ReproBAIT, an evaluation with SecureBio measuring whether AI coding agents can reproduce existing biological AI tools from limited public information. Agents were given a paper or abstract, a fixed time and compute budget, and an isolated sandbox. The simplest tools could be partly reproduced within 4 hours for under $20, while the most complex tool, RFdiffusion3, was not successfully reproduced. Results suggest withholding code does not alone prevent reconstruction of selected tool functions.
One primary source; other links are context, not independent corroboration.
- Primary sourceX discovery · source check open
- Discovery linkX
- Linked in sourceSecureBio
- Linked in sourceProteinMPNN
- Linked in sourceRFdiffusion3
- Linked in sourcehere
- Linked in sourcehere
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Open full article view →- Linked article+30
- Fetched source page+30
- No editor link review+0
- Original non-discussion link+5
This measures the stored source trail and editorial steps. It does not certify accuracy, peer review, or independent corroboration. Popularity adds no points.
View full scoring →Read the scoring rubric →What the source says
The UK AISI Chem-Bio team built ReproBAIT, an evaluation with SecureBio measuring whether AI coding agents can reproduce existing biological AI tools from limited public information. Agents were given a paper or abstract, a fixed time and compute budget, and an isolated sandbox. The simplest tools could be partly reproduced within 4 hours for under $20, while the most complex tool, RFdiffusion3, was not successfully reproduced. Results suggest withholding code does not alone prevent reconstruction of selected tool functions.
Key findings from the stored brief
- Agents reproduced selected functions of the simplest tools within 4 hours and under $20 in compute.
- ProteinMPNN reached 96.5% reproduction fidelity with Claude Opus 5 given the full paper.
- No agent successfully reproduced RFdiffusion3 within its time budget, with attempts costing $102 to $301.
- Withholding code alone does not equate to a safe release, pointing to data governance as a possible mitigation.
Reproductions cover only selected functions of each tool, and memorisation of public papers and code by the agents could not be ruled out, so the source's diagnostics are described as directional rather than conclusive.
This is a source-based abstract, not independent validation of the source's claims.
Evidence & sources
These links show where the signal came from and what the source page linked to. A linked page is context, not independent corroboration.
Related archive coverage
These relationships record a citation or topic overlap, not confirmation of the finding.
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Timeline & updates
- Signal dated by source or discovery feed
This timestamp places the signal in the radar. It may be the discovery post date rather than the primary page's first publication date.
- Shared brief stored
AINewsIQ prepared one abstract from fetched source page.
A publication revision history is not stored for this signal. New engagement snapshots do not mean the source itself changed.
Why this may matter
The findings bear on whether restricting code or detail in dual-use biological AI tools is an effective safeguard, and suggest data governance may need more attention as agent capabilities improve.
Limits and open questions
Reproductions cover only selected functions of each tool, and memorisation of public papers and code by the agents could not be ruled out, so the source's diagnostics are described as directional rather than conclusive.
Organization-specific financial, security, and operational effects require the organization's own context and review.
Discussion
Ask about the source, add context, or point readers to a relevant follow-up.
Discussion
Comments are reviewed before publication. Discuss the evidence, not the person.
No published comments yet.
Reader discussion opens with accounts. You can read the original source and published comments now.