Stephen Herzog
@herzogsm.bsky.social
250 followers 96 following 46 posts
Professor of the Practice, James Martin Center for Nonproliferation Studies | Associate, Harvard Project on Managing the Atom | nuclear arms control, nonproliferation, disarmament, technology
Posts Media Videos Starter Packs
Reposted by Stephen Herzog
liviuhorovitz.bsky.social
SATCOM & Bundeswehr, Russia’s upgraded missiles, Iran's nuclear program, DPRK & South Korea, AI & tech race & more in this week’s STAND newsletter, a project by @swp-intsecurity.bsky.social, on recent news & analyses to better #underSTAND #nuclear #space #deterrence & #threats 1/10
Reposted by Stephen Herzog
herzogsm.bsky.social
7/7
Our goal is not to predict the future, but to promote thinking about how emerging technologies affect proliferation outcomes. Follow the link to the simulation tool in the article to test your own model values/assumptions & observe how proliferation & verification interact.
herzogsm.bsky.social
6/7
We find that early investments in advanced detection tech have big effects in offsetting the potential for AI-enhanced nuclear breakout.
herzogsm.bsky.social
5/7
To capture this uncertainty we simulate 6 scenarios. They range from AI having only limited proliferation relevance to AI replicating nuclear weapon design teams. We also model detection's current innovation pace & compare to “moonshot” investments in monitoring/verification.
herzogsm.bsky.social
4/7
Detection leads, but verification has historically evolved step-wise, often in response to crises. Commercially-driven dual-use technologies with WMD applications are currently evolving faster, with uncertain endpoints. Could we run into the situation depicted in this plot?
herzogsm.bsky.social
3/7
Think of nuclear breakout as a race between would-be proliferators & actors engaged in monitoring & verification. We model this race: AI-aided trajectories for proliferation-enabling vs. detection-enhancing technologies.
herzogsm.bsky.social
2/7
States will need a lot more than a ChatGPT subscription to build a bomb. But we examine how growing AI capabilities may reduce breakout time & change latent proliferators’ calculations about their likelihood of covert success.
herzogsm.bsky.social
Atomic Backfires: When Nuclear Policies Fail is now available for pre-order from @mitpress.bsky.social! The thread below describes the premise of the volume and its contributors. mitpressbookstore.mit.edu/book/9780262...
Reposted by Stephen Herzog
herzogsm.bsky.social
1/4
Congratulations to Nonproliferation Review McElvany Award winners for their articles! @lydiawachs.bsky.social (grand prize), @karldewey1.bsky.social (runner-up), & Austin Cooper (honorable mention). Links to #openaccess articles in subsequent posts below.
nonproliferation.org/article-on-r...
Article on Russian nuclear strategy during Ukraine war takes top McElvany Award
An article providing key insights into Russian nuclear strategy during the first 18 months of the war in Ukraine has won the grand prize.
nonproliferation.org
herzogsm.bsky.social
Was delighted to be at KAIST yesterday for the U.S.-Korea International Workshop on Research Security and Integrity. We had very productive conversations about safeguarding research ecosystems against R&D misappropriation with incredibly welcoming and knowledgeable hosts!
Reposted by Stephen Herzog
matthew-bunn.bsky.social
Thoughtful analysis of the perspectives from Europe, Russia, and Iran, and of a narrow potential path to accord from @nicolegrajewski.bsky.social. In addition, she's been posting a number of useful items on action in Iran's parliament, the text of Russia's draft UNSC resolution, etc.
herzogsm.bsky.social
It's even more fun when LLMs are generous enough to invent fictitious co-authors of the article you never wrote.
kristianulrichsen.bsky.social
I'm peer reviewing a manuscript for a reputable publisher and came across fake references which must be AI-generated - including to an article attributed to me which sounds fascinating but which I never wrote.
Reposted by Stephen Herzog
herzogsm.bsky.social
1/13
In a new report for @vertic.bsky.social, I take a hard look at the challenge that highly latent nuclear states present to irreversibility in nuclear disarmament. Summary and relevant tables and figures below. Read the report here: www.vertic.org/wp-content/u...
herzogsm.bsky.social
13/13
I address these questions and more in the report! www.vertic.org/wp-content/u...
www.vertic.org
herzogsm.bsky.social
12/13
How much confidence is enough, and will states be willing to pay for it? And which states should be the ones to do so?
herzogsm.bsky.social
11/13
And there is a trade-off between verification success and costs, as the former asymptotically approaches 100% at increasingly greater expense but never quite gets there.
herzogsm.bsky.social
10/13
This will, of course, be expensive and is also a matter of political will. But the framework is, in my view, feasible to implement.