iiinova — where your digital self keeps evolving

Cas (Stephen Casper) @scasper.bsky.social · 13/10/2026

Harvard SEAS has a search out for a tenure-track CS prof this year. You can email me if you have any questions.

Cas (Stephen Casper) @scasper.bsky.social · 09/10/2026

I am only getting to read this 3 weeks after it was published, but I was very happy to see this in China's CAC's AI Safety Governance Framework 3.0. Exhibit N on how the lack of AI gov collab between the US and China is more the US's fault than China's. www.cac.gov.cn/rootimages/...

110

Cas (Stephen Casper) @scasper.bsky.social · 25/09/2026

🧵 I read the Sanders/Casar "Ban Artificial Superintelligence Act of 2026" I wouldn't pass this in its current form, but there are some things I like about it. Thoughts here in thread.

100

Cas (Stephen Casper) @scasper.bsky.social · 28/08/2026

Yesterday, Robert Wright and I discussed what will happen when AI systems become an invasive, parasitic species in cyberspace that undergo digital and cultural evolution.

250

Cas (Stephen Casper) @scasper.bsky.social · 10/08/2026

Glad to see this from @sanders.senate.gov. www.sanders.senate.gov/wp-content/u...

2114

Cas (Stephen Casper) @scasper.bsky.social · 05/08/2026

None of the 19 incidents that UK AISI found were from 'helpful-only' models. There is a case to be made that setting Mythos- or GPT 5.6-Sol-level AI cyberagents to run without a robust real-time monitoring setup is an inherently (perhaps abnormally) dangerous activity.

010

Cas (Stephen Casper) @scasper.bsky.social · 28/07/2026

With open models, safeguards can and will be stripped away. For example, there are over 13,000 models on Hugging Face that have been downloaded, stripped of external safeguards (e.g. safety filters), trained to comply with any request, and then re-uploaded.

100

Cas (Stephen Casper) @scasper.bsky.social · 28/07/2026

In practice, though, it's challenging to make open models safe against sophisticated attempts at misuse. Deployers control all points of access to closed models, and they have a lot of layers of Swiss cheese they can stack.

200

Cas (Stephen Casper) @scasper.bsky.social · 28/07/2026

First, let's be clear: open-weight models are simultaneously wonderful and terrible. They diffuse power and enable safety research. I struggle to imagine a positive AI future without these things at play. But they also enable tremendously harmful forms of unchecked misuse.

100

Cas (Stephen Casper) @scasper.bsky.social · 28/07/2026

🧵 If trends hold, expect a Mythos-level open-weight model around Christmas. Meanwhile, open models are important but also a massive hole in most agendas for safe AI. Here's a 🧵 of my thoughts & research agenda on the technical & political challenges we need to address.

1112

Cas (Stephen Casper) @scasper.bsky.social · 25/07/2026

I was wondering if there was any research about how AI companies sometimes perversely keep their safety research to themselves to build a moat around it and gain a competitive edge over competitors. I found one. It's a cool paper. Sharing here in case anyone's interested.

210

Cas (Stephen Casper) @scasper.bsky.social · 24/07/2026

Meanwhile, AI politics conversations empirically seem to be very incident-driven. For example, note how suicide coaching and mass nudification incidents in the past year have led to an avalanche of lawmaker interest in recent months.

100

Cas (Stephen Casper) @scasper.bsky.social · 24/07/2026

🧵 In AI, we are used to seeing graphs that start to exhibit hockey stick behavior around 2023-2025. But that's a little bit funny and incongruous in light of how relatively little the Overton window has changed with AI lawmaking since 2024...

110

Cas (Stephen Casper) @scasper.bsky.social · 23/07/2026

The summary released today of the FRONTIER Act is cool. It seems like a pretty rigorous bill. Based on the summary, in my opinion, it might be good enough to be worth passing. But I would still tweak a few things. Here is a brainstorm of 8 ideas. 🧵

111

Cas (Stephen Casper) @scasper.bsky.social · 23/07/2026

🧵 Mini book review: The Chicago School by Johan Van Overtveldt

100

Cas (Stephen Casper) @scasper.bsky.social · 22/07/2026

Want to get up to speed on what researchers have been saying about internal deployment of AI lately? I would recommend checking out these four papers. Let me know in the replies if I'm missing something.

110

Cas (Stephen Casper) @scasper.bsky.social · 22/07/2026

OpenAI's internally deployed models hacking Hugging Face does not seem to have been unpredictable or inevitable. We talked about the root of the problem & what policymakers can do about it back in February. Props to Joe for hitting the nail on the head.

152

Cas (Stephen Casper) @scasper.bsky.social · 20/07/2026

This leaves one major question. What causes this? Why does this only happen with some companies? We have three hypotheses: 1. It's intentional and by design. 2. It's predictable but unintentional. 3. It's emergent misalignment.

250

Cas (Stephen Casper) @scasper.bsky.social · 20/07/2026

In an informal poll we sent out over our socials, the crowd was ok at guessing some of the results, but had a big blind spot for DeepSeek.

120

Cas (Stephen Casper) @scasper.bsky.social · 20/07/2026

xAI is the worst, followed by DeepSeek, Anthropic, and OpenAI.

261

Cas (Stephen Casper) @scasper.bsky.social · 20/07/2026

We tested 7 hypotheses, one for each company. xAI, DeepSeek, Anthropic, and OpenAI models all clearly have a tendency to downplay company controversies. p<10^-5 (our permutation tests bottomed out). Google, Meta, and Alibaba models don't seem to do this, though!

120

Cas (Stephen Casper) @scasper.bsky.social · 20/07/2026

We elicited open-ended discussions from 21 models from 7 companies on 206 negative news stories using 25 prompt templates to assess how favorably each model discusses controversies from each company.

120

Cas (Stephen Casper) @scasper.bsky.social · 20/07/2026

AI models are major mediators of information. Recognizing this, many AI companies have publicly discussed the importance of AI systems not taking positions or disseminating information in ways that favor special interests. Do they mean it?

120

Cas (Stephen Casper) @scasper.bsky.social · 20/07/2026

🚨 New paper: Some, but not all, AI companies make corporately-loyal models. xAI, DeepSeek, Anthropic, & OpenAI models all downplay company controversies. Google, Meta, & Alibaba models don't. The findings are clear, but we are pretty confused as to why... 🧵 @finke.dev

2186

Cas (Stephen Casper) @scasper.bsky.social · 09/07/2026

Stability is now being sued (alongside xAI) for abetting the production of AI NCII/CSAM due to how it developed & released several open-weight models. Anyone interested in whether AI companies will be held liable for foreseeable, mitigatable *downstream* harms should follow this.

172

Cas (Stephen Casper) @scasper.bsky.social · 03/07/2026

162 responses so far. More uniform spread than I expected. Only 4 have been right (slightly worse than random chance).

000

Cas (Stephen Casper) @scasper.bsky.social · 03/07/2026

Just saw this new paper. It was already known that models from Stability and Alibaba dominate the image & video NCII ecosystems, respectively, but I didn't know they were *this* dominant. Just a few socially reckless companies are the principal enablers of AI NCII abuse.

240

Cas (Stephen Casper) @scasper.bsky.social · 02/07/2026

Responses so far... (Also, Hi Ben, haven't talked much since LS50 -- I'm back at Harvard though!)

210

Cas (Stephen Casper) @scasper.bsky.social · 02/07/2026

Lennart Finke and I will release a paper on Monday about how some AI developers tend to make models that differentially downplay company controversies. Below (🧵) is a link to a 1-question Google form for you to guess the results before they're out. (They might surprise you.)

2111

Cas (Stephen Casper) @scasper.bsky.social · 22/06/2026

I am starting an AI governance ICML megachat on Signal. DM me if you'd like the link to join.

151

Cas (Stephen Casper) @scasper.bsky.social · 20/06/2026

Commission-appointed board members would not have any overriding duties to the company's financial interests -- only to Americans.

110

Cas (Stephen Casper) @scasper.bsky.social · 20/06/2026

The second major power of the Commission is to appoint board members for companies that are part of the sovereign wealth fund. Since the fund owns 50% of equity/capital, this basically means up to 50% membership control over all AI company boards.

100

Cas (Stephen Casper) @scasper.bsky.social · 20/06/2026

However, an industry bailout is expressly prohibited.

100

Cas (Stephen Casper) @scasper.bsky.social · 20/06/2026

The Commission is given the power to make payments from this fund every year up to 5% its average value of the prior year. The Act says that it will be used to make direct payments to Americans and, more vaguely, to ensure that every American has a decent standard of living.

100

Cas (Stephen Casper) @scasper.bsky.social · 20/06/2026

There are some additional details about (1) how, if your company qualifies, it may need to reorganize to structurally separate its AI business from its non-AI business, and (2) closing some loopholes involving companies organized outside of the US.

100

Cas (Stephen Casper) @scasper.bsky.social · 20/06/2026

How do you know if your company qualifies? The basic idea is that you qualify if you spend $200 million or more within any calendar year on AI datacenters, computing infra, services, or advanced robotics. (Each has a specific def. not shown in screenshot.)

100

Cas (Stephen Casper) @scasper.bsky.social · 20/06/2026

The first major role of the Commission is to manage payments from the new sovereign wealth fund. This Act taxes 50% of the stock and capital of qualifying AI companies.

100

Cas (Stephen Casper) @scasper.bsky.social · 20/06/2026

That Commission would consist of 7 members with 5-year terms and some fairly commonsense requirements around conflicts of interest. No more than 4 Commissioners can be from the same party. They are nominated by the President and confirmed by the Senate.

100

Cas (Stephen Casper) @scasper.bsky.social · 20/06/2026

The bill establishes a commission within the Treasury Department known as the "Independent Commission for Democratic AI".

100

Cas (Stephen Casper) @scasper.bsky.social · 20/06/2026

I just gave Bernie's AI Wealth Fund Act a close read. The wealth redistribution via this Act would be enormous. But it does something else far more impactful... 🧵Here's what it does, plus 3 things I'd change -- one of which I think, if unaddressed, could be a fatal flaw.

220

Cas (Stephen Casper) @scasper.bsky.social · 15/06/2026

A generationally important US House primary for AI and tech policy is happening on June 23 in New York's 12th district. If you live in NY-12, and if you believe that AI safeguards, transparency, and accountability are critical, I hope you consider voting for @alexbores.nyc (D).

2112

Cas (Stephen Casper) @scasper.bsky.social · 12/06/2026

Now that I have your attention by posting this spinning point cloud GIF, I'd like to propose a litmus test for AI mechanistic interpretability research. You might call it the "interp hammer" test...🧵

160

Cas (Stephen Casper) @scasper.bsky.social · 09/06/2026

Anthropic and OpenAI are publicly pointing out how having the option to slow down AI would offer a potentially critical form of optionality in the future. The correct response for any policymaker should be "Damn, this is serious. How can I help build that capacity?"

040

Cas (Stephen Casper) @scasper.bsky.social · 08/06/2026

According to the MIT Libraries' database of theses (dating back to the 1800s), my thesis was only the 2nd in the institute's history to contain the word "shit."

2170

Cas (Stephen Casper) @scasper.bsky.social · 05/06/2026

Sometimes I run into old papers or books that I can't believe weren't written about AI today.

050

Cas (Stephen Casper) @scasper.bsky.social · 01/06/2026

Just finished my PhD at @MITCSAIL. In July, I'll start as an assistant professor at the @harvard.edu @harvardkennedy.bsky.social. I have lots to learn and lots to do. With others (some TBA 👀) at HKS, I'm looking forward to helping academia offer guidance for governing the next chapters of AI.

3170

Cas (Stephen Casper) @scasper.bsky.social · 28/05/2026

As a personal exercise, I drafted a proposal for how I would amend section 230 -- it would only apply to platforms that are small or have open-source recommendation algs. (This passes the Wikipedia test.) Note: I'm not the first person to have this idea. Just sharing for fun.

030

Cas (Stephen Casper) @scasper.bsky.social · 23/05/2026

Thank-yous from my PhD thesis defense earlier this month.

030

Cas (Stephen Casper) @scasper.bsky.social · 14/05/2026

In practice, almost all AI porn is, in a certain morally relevant sense, not completely consensual. I was glad to talk briefly about this with Jessica Klein while she was writing this new MIT Technology Review piece!

100