Modular pretraining lets dangerous capabilities be switched off per module
Confirmed
The takeaway
On July 8, 2026 researchers from AE Studio and Anthropic published “Modular Pretraining Enables Access Control”.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 2 of 5
- Last verified
- 1 October 2026
Your AI and this story
- GPT-6.2 Astra38 days after its cutoff
- Claude Opus 5.58 days after its cutoff
- Gemini 4 Hydrogen99 days after its cutoff
- Grok 4.738 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 8 days before it.
Key facts
- 800M-parameter model with four dual-use domains: GRAM and LoRA matched data filtering at capability removal and beat it on retained domains
- Capability removal improved with scale across 50M–5B parameters
- Four modules give 16 configurations from one training run, vs one run per configuration for data filtering
- With only 50% of data labelled, GRAM isolated capabilities better than filtering or LoRA
- Authors include Cem Anil and Alex Cloud (Anthropic) and Judd Rosenblatt (AE Studio)
What happened
An extension of gradient routing. Dangerous knowledge is placed in removable modules during pretraining instead of being filtered out of the data.
Why it matters
It points to tiered access, where vetted users get a model with, say, the virology module and the public does not, without training separate models. This bears on how labs might ship open or differentiated weights safely.
Sources
1 source from 1 site. Numbers match the chips in the text.
1 source: 1 primary
Primary
- Anthropic Alignment Science blog: Modular Pretraining Enables Access Controlalignment.anthropic.com, paper
Changes
- Filed