Alex Engler

I learned a lot from @bankston.bsky.social's long post on the debate over the AI Preferences technical standard, and strongly recommend it. It is among the most essential issues for the future of the internet and access to information. cdt.org/insights/wil...

Will the Open Internet Survive the War on Bots? Mapping the Debate Over “AI Preferences” - Center for Democracy and Technology cdt.org

In the @inquirer.com, @geurkink.bsky.social clearly lays out the unambiguous reality that the Trump administration is engaging in a censorship campaign against researchers of technology companies and the information ecosystem. www.inquirer.com/opinion/comm...

The U.S. is targeting the researchers who help us understand the effects of social media | Opinion inquirer.com

Generally agree! In your thinking, does the "how" that compute is used matter for it to be a reasoning model? For instance, this paper makes claims about what methods count as reasoning (tho I'm not sure I buy this) arxiv.org/pdf/2501.11223

arxiv.org

I just learned about ORES at a conference on Friday and think it's fantastic. I'm reading what you sent over + also just found this preprint (Bye Bye Perspective API) arxiv.org/pdf/2604.25580

arxiv.org

Totally surreal to announce I've been targeted by my first China-aligned threat actor, as an AI policy leader. Sometimes little moments of recognition like this really make you feel like you've made it www.reuters.com/legal/govern...

reuters.com

Also, I would note politely to @jonathanvanian.bsky.social, who wrote the letter up for CNBC, that the letter does not mention the word "safety." Independent evaluations are necessary for a wide range of pro-social outcomes, far beyond a narrower "safety" framing. www.cnbc.com…

Anthropic and OpenAI need truly independent safety evaluators, experts say in public letter cnbc.com

SAGA is a meaningful step in this direction, recognizing we need to broadly empower sectoral regulators to govern AI. SAGA's text would benefit from more specific authorities (see my paper or oped) but I’m glad to see this bill has gone through the House hopper www.techpolicy.…

Proposing the CASC: A Comprehensive and Distributed Approach to AI Regulation techpolicy.press

My takeaway is we need sectoral governance of AI, where agencies have the domain knowledge, technical expertise, digital infrastructure, and regulatory authorities to effectively govern AI systems within their domains. Here's the original @brookings.edu paper: www.brookings.ed…

A comprehensive and distributed approach to AI regulation | Brookings brookings.edu

In my first piece in 3 years (!), Danaé Metaxa and I write for @carnegieendowment.org on why we need a new paradigm of scientific research for AI—longitudinal monitoring—to ensure the public has a meaningful say in the societal role of LLMs. carnegieendowment.org/research/202...

We Need Ongoing Monitoring of AI and Political Information carnegieendowment.org

Great paper from Duncan Watts: LLM-based content reranking improves personalization & aligns video recommendations more closely to users’ prior viewing. Very interesting tho unclear if there is significant deployment of LLM-based recommenders systems at present? infodem.upenn.…

Research Compendium - Center on Media, Technology and Democracy infodem.upenn.edu

The attackers posed as former OSTP official Lynne Parker with a believable email about a new AI policy center, which was apparently oriented towards U.S. AI policymakers - be careful out there! Thanks to @proofpoint.com for the thorough technical investigation www.proofpoint.c…

Hallucinating Credibility: China-Aligned TA419 Impersonates its Way into US AI Policy Circles | Proofpoint US proofpoint.com

I am a signatory of this letter calling for researchers and evaluators to have access to AI models, with meaningful guarantees of their independence, as an essential transparency and accountability mechanism. aievaluatorforum.org/initiatives/...

AI Evaluator Forum aievaluatorforum.org

@penn-mediated.bsky.social is at the #CJ2026 and would love to chat with folks. Come to our workshop where you can run your own LLM audit (today at 3:30) or I'm giving a talk tomorrow on the broader needs for scientific and longitudinal monitoring of LLM politics (in Parallel Talks 1 at 1:30).

I think it's true that AI has important implications for national security. However, the overwhelming majority of people who make this argument are making underspecified claims about how and in what specific systems, making most of the related analysis or conclusions useless.

Writing about automated content moderation - it's amazing how many papers were focused on toxicity, with the comparison point being the Perspective API The sunsetting of the API means a loss for reproducibility And, this will probably be the case for many papers based on commercial LLMs

I just finished the first real draft (ready for reviewers) of a paper since I published on AI in housing valuations in October 2023 (the day I joined Biden admin) Was a real battle to get back to research/writing but it feels good now