Modular Pretraining Enables Access Control
Anthropic has introduced GRAM, a new technology designed to provide an 'off switch' for isolating and managing specific knowledge within AI models.
Velocity timeline
How fast coverage is spreading — measured hourly from article rate × source diversity. How this works →
The brief
Anthropic has unveiled GRAM, a technology aimed at modularizing AI pretraining to control access to specific dual-use knowledge. The system is designed to isolate potentially dangerous information within a model without impacting its broader capabilities.
Coverage from Anthropic, Dataconomy, and finance.biggo.com emphasizes the technical function of the tool as an 'off switch' for AI knowledge. The Telugu Times and Alignment Science Blog highlight the implications for managing safety and preventing potential misuse in AI development.
Future developments will focus on the practical integration of GRAM into existing model architectures. Reporting does not yet specify how the technology will be deployed or whether it will be extended to external developer platforms.
Synthesized by Newsylist from the headlines below under a strict no-invention contract. ✓ fact-checked: all claims supported by sources Updated just now.
Quick answers
What is GRAM?
GRAM is a technology introduced by Anthropic designed to isolate and provide an 'off switch' for dual-use knowledge in AI models.
What is the intended purpose of this technology?
It is intended to enable access control over specific AI knowledge to prevent potential misuse.
Does this tool delete information from the AI?
Coverage describes the technology as an 'off switch' for isolating knowledge, rather than the permanent deletion of an AI's overall processing capabilities.
Coverage (5)
- Anthropic Unveils 'Off Switch' for AI Knowledge with New GRAM Technology to Prevent Misuse finance.biggo.com · 1d ago
- Anthropic Research Introduces GRAM For Isolating Dangerous AI Knowledge Dataconomy · 1d ago
- A ‘Delete Button’ for the AI Brain? If AI Knows These Things, Could It Become Dangerous? Telugu Times · 1d ago
- An off switch for dual-use knowledge in AI models Anthropic · 1d ago
- Modular Pretraining Enables Access Control Alignment Science Blog · 1d ago broke it first
People, places & organizations
Topics
Related trends
Leap Seconds Were a Headache. Now Timekeepers Are Considering a Leap Hour
International timekeepers are preparing to vote on a significant shift in how global time is synchronized: moving from leap seconds to a potential leap hour.
OpenAI unveils long-awaited "super app" as rivalry with Anthropic intensifies
4 news sources are covering this Business story right now — Newsylist is tracking how fast it spreads.
NVIDIA announces 14 new GeForce cards, but none of them have a GPU
5 news sources are covering this Technology story right now — Newsylist is tracking how fast it spreads.
iPhone 18 Pro Max component costs could jump by nearly $300, per report
Reports indicate component costs for the upcoming iPhone 18 Pro Max may increase by nearly $300 per unit.
I swapped my smartphone for a 'dumb phone'. It changed me
5 news sources are covering this Technology story right now — Newsylist is tracking how fast it spreads.
Anthropic 3Q26 Profit Over $1B: The Anthropic IPO Financials Sneak Peak
Anthropic reports third-quarter profits exceeding $1 billion ahead of a highly anticipated public offering.