Mia Hoffmann

Research Fellow @CSETGeorgetown focusing on AI harms and assessment | former researcher @Bruegel_org examining AI and the future of work. Views are my own.

[CSET @CSETGeorgetown]

In a new report, @KyauMill21, @mia_hoffmann_, and Rebecca Gelles investigate the use of open-weight AI models by researchers. They review ~250 scientific publications, identify use cases that are enabled by access to open weights, and consider the broader impact on research.

[CSET @CSETGeorgetown]

✨New blog✨
Red-teaming is an essential testing method for AI safety and security, but it's hard to get right.
In this new blog, Evelyn Yee discusses threat modeling and how evaluation tools can make or break a red-teaming exercise. https://t.co/NOgTsiEjxa

[Cole McFaul @colemcfaul]

Really interesting report here on open models by my colleagues @KyauMill21 and @mia_hoffmann_!

Norms are underrated. And open-source is a cultural feature of & software development. PRC open-source models generally outperform their US counterparts.

I think that matters 🧵/7

[Kyle Miller @KyauMill21]

🧵Most agree that open-weight AI models benefit research, but there is a lack of empirical work investigating how these benefits manifest in practice. To address this gap, we reviewed ~250 scientific publications that use open models in ways that require access to the weights

[CSET @CSETGeorgetown]

✨New Report✨

AI guidance is scattered, technical, and often inaccessible. Deciphering the myriad of recommended practices can be overwhelming. We created a harmonized framework—so organizations don’t have to untangle 7,000+ recommendations.

Read more: https://t.co/4lxWB0erzs

[Cole McFaul @colemcfaul]

🚨 NEW REPORT! 🚨

After 2 (!!) years working on a new dataset of thousands of PLA contracts, I’m very excited to share a new report, featured in today’s @WSJ!

We ask: Who are the PLA’s AI suppliers?

The answer: an unexpected, emerging class of firms and universities.
🧵/15

[Sam Bresnick @SamBresnick]

1/ 🚨 In a new @CSETGeorgetown report, @colemcfaul, Daniel Chou, and I examine the inner workings of China’s military-civil fusion strategy and AI defense industrial base. Here’s what we found 🧵

[CSET @CSETGeorgetown]

���New Data Snapshot

@NIH funding helps fuel innovation directly — resulting in 84,000 U.S. patents — but its impact is felt much more broadly.

Almost half of all research areas in the global patent landscape contain NIH-funded research.

[CSET @CSETGeorgetown]

📢 CSET is hiring!

We’re looking for a Data Research Analyst to produce data-driven research products and policy analysis as a member of CSET’s data team!

Interested or know someone who would be? Learn more and apply by September 2 👇 https://t.co/OidjbrsCYY

[CSET @CSETGeorgetown]

Can the new EU AI Code of Practice change the global AI safety landscape?

As companies like Anthropic, OpenAI, and Google sign on, CSET’s @mia_hoffmann_ explores the code’s Safety and Security chapter: https://t.co/cXkJIVpmC3

[Cole McFaul @colemcfaul]

Important document for understanding US AI foreign policy.

Three takeaways:
1. Not a zero sum outlook on S&T diplomacy! 👍
2. First time I've seen the US govt use "Sovereign AI" 👍
3. Kratsios says "AI race has no finish line." I agree, but does the rest of the WH?🧵
1/8

[Sam Bresnick @SamBresnick]

In @NikkeiAsia, I argue that, in contrast to their support for Ukraine in 2022, tech companies may hesitate to support Taipei in a crisis due to their economic exposure to Beijing. Unlike Russia, China is central to many of these companies’ supply chains and strategic ambitions.

[CSET @CSETGeorgetown]

🇪🇺New Analysis🇪🇺

Can the new EU AI Code of Practice change the global AI safety landscape?

As companies like Anthropic, OpenAI, and Google sign on, CSET’s @mia_hoffmann_ explores the code’s Safety and Security chapter. (Link in 🧵)

[Mia Hoffmann @mia_hoffmann_]

Read the full post here: https://t.co/UrHViwP71L

[Mia Hoffmann @mia_hoffmann_]

Could the EU’s AI Code of Practice set a new global standard for AI safety?
In my new blog post, I break down the safety & security chapter and analyze its potential to shape frontier AI governance for the better.
🧵

[Mia Hoffmann @mia_hoffmann_]

Even if companies develop their own practices to comply with the AI Act, the Code is significant because it's a clear articulation of the depth and rigor the EU expects of frontier AI developers when addressing systemic risks. As such, it already sets a new standard for AI safety