logan koepke

“In light of the above, the Commission believes that AI companies that steer the outputs of their AI systems toward unexpected objectives, and away from the objectives set by or reasonably expected by users, are likely to deceive.” www.ftc.gov/news-events/...

FTC Seeks Public Comment on Policy Statement Addressing AI Accuracy ftc.gov

🧵 How do companies in highly regulated domains actually test their AI systems for and mitigate discrimination? Our new @upturn.org paper at @facct.bsky.social w/ Emily Black, @mbogen.bsky.social, @s010n.bsky.social, and @wesleydeng.bsky.social, tries to answer this question: a…

The Fair Lending Model: How the Longest-Running Algorithmic Fairness Programs Work in Practice arxiv.org

hm. I don’t see my question answered in this piece. I agree independent evaluation is more than necessary, but to me the relevant question is if IVOs would have verified the labs followed a standard given their practices during the relevant time period. fortune.com/2026/08/08/…

AI labs shouldn't be allowed to grade their own homework | Fortune fortune.com

yes, and — even worse: today AI systems today enforce *existing austerity* of our public benefits systems and social safety net, helping facilitate determinations of who is worthy and not of care, food, unemployment, etc. only made worse by OBBBA's Medicaid cuts + work require…

Benefits Tech Advocacy Hub btah.org

quite interesting design transluce.org/announcing-m...

Announcing Transluce's Mental Health Evaluation transluce.org

my 💭 to the trahan-obernolte AI discussion draft conversation: if the stated motivation is preempting state laws re: *frontier model development,* why does the text preempt "AI model" development and not "frontier model" development, given the definitions are different in the bill's text?

very strange to observe the policy debate about ai auditing basically cycle through the debate in FAccT papers about the ecosystem and institutional design of ai auditing from 2017-2022, but functionally give zero acknowledgment of this literature or work.

the slaughter scotus decision basically imperils any ai governance policy proposal that would rely upon multi-member, independent commissioners to do technical licensing calls.

theory of change: befriend someone in middle school who has a father who is elementary school friends with the person who becomes the Pope, talk about AI

given rapid new interest in the material effects of ai systems, i think it is extremely unfortunate that many on the left are not seen as starting where people are on ai issues.

noticeable that the other ai labs, and seemingly a lot of the research community, are silent as xAI and the DOJ mount an effort to declare a law requiring testing AI systems for discrimination is unconstitutional.

10 year anniversary of this White House AI report in a few weeks

personally, if you think ai systems will be important at the level of economy-defining scale, private governance will not be sufficient given the amount of risk *assumed by the public.* you need robust state capacity to manage public risk.

guess OpenAI's stated "our contract is keyed to laws as they exist today" red line w/ the DOD will get a real workout with the new NSPM directing the DOD to update 3000.09 in 90 days and then reviewed annually.

imagine it's January 2026 and an independent verification organization is tasked with independently verifying whether a frontier AI lab met a cybersecurity standard. would it likely have certified that OpenAI or Anthropic met the standard? if yes, what does that suggest?