250 Years of Legislative Legacy
Today’s case for legislative data and the legacy you are contributing to.
This is the July edition of Keeping Pace, formerly Future-Proofing Congress. POPVOX Foundation’s mission is to help Congress close the gap between how fast technology is moving and how well institutions understand it. We call this the Pacing Problem. Keeping Pace hits inboxes monthly, giving Congressional and legislative staff across institutions practical, nonpartisan AI and technology education: no vendor bias, no asks, no politics. Learn more and subscribe here.
Newsletter length: This is a 10 minute read.
LETTER FROM THE EDITOR
Today, the Committee on House Administration Subcommittee on Modernization and Innovation is hosting a hearing on “Modernizing Public Access to Legislative Data and Information,” and it could not come at a more important time. As we celebrate America’s semiquincentennial, we are reminded that the American government (Congress included) is a treasure trove of our heritage. Our nation’s legislative legacy is the debates, bills, laws, reports, records, and personal contributions of the citizens, Members, and staff who have participated. Your work is a growing contribution to that collection.
Reliable public access to data from trusted sources (such as Congressional and other Legislative branch agency websites) isn't just important for humans anymore. As individuals increasingly rely on AI tools to find information and get access to services, the ability for public data to be ingested by machines is a prerequisite for ensuring their accuracy and the discoverability of Congress’ information (like how a Member voted or what legislation they have sponsored).
Not all data is created equal, and modernizing access no longer means simply scanning a paper document to have the PDF available. Lasting, substantive modernization means turning that PDF into strong, clean metadata that both humans and machines can easily find and ingest. The challenge for Congress (and all legislatures around the world striving to increase access to information) is a dynamic one: modernizing public access to legislative data will not succeed if the sole focus is on how to make resources more accessible today. Institutional change is required to ensure that legislative information is distilled and structured so that it is future-proofed, and more accessible for our next 250 years.
The Committee on House Administration hearing begins at 2 PM ET.
This month, we’re highlighting:
The dynamics at play around the release of powerful new frontier models and the White House’s in-depth involvement in controlling access to them
Tips to avoid LLM lock-in and minimize transition pain if your office switches from one institution-approved LLM to another
How to use an LLM as a House and Senate Rules tutor
How AI is democratizing app creation and Congressional information
Some additional events, reports, AI tools, and items that caught our attention
Aubrey Wilson
Managing Director
POPVOX Foundation
TECH WATCH
Mythos, Fable, Cyber, Sol and the question of control
In our May newsletter, we discussed Claude Mythos, an Anthropic AI model so effective at finding cybersecurity vulnerabilities that the company restricted access to vetted partners through Project Glasswing. In June, Anthropic released Fable 5, a public-facing version of Mythos. On June 12, after only days of being available, the Commerce Department placed both Mythos and Fable under export controls due to national security concerns, and Anthropic suspended customer access entirely, both domestic and international. Limited access to Mythos was re-authorized by the White House on Friday, June 26. Fable remains offline.
On June 22, OpenAI released an update to its own cybersecurity-focused model, GPT-5.5-Cyber. On CyberGym, an internal benchmark that measures whether an AI agent can reproduce known software vulnerabilities, the updated model scored 85.6%, slightly ahead of Mythos 5’s 83.8% on the same test. OpenAI says GPT-5.5-Cyber does not have built-in safety guardrails, meaning it can be used for red-teaming and penetration testing. As a result, OpenAI has restricted its access to organizations that are part of its cybersecurity-focused Daybreak initiative. On June 26, OpenAI also announced that their newest frontier model suite — ChatGPT-5.6 Sol, Terra, and Luna — is being released on a user-by-user basis, upon approval by the White House.
From OpenAI’s announcement:
We don’t believe this kind of government access process should become the long-term default. It keeps the best tools from users, developers, enterprises, cyber defenders, and global partners who need them. We are taking this short-term step because we believe it is the strongest path to broader availability in the coming weeks, while we work with the Administration to develop the cyber Executive Order framework and a repeatable process for future model releases.
Screenshot of graphic accompanying OpenAI’s announcement of the ChatGPT 5.6 model suite
For Congressional staff, the relevant point is that these frontier models are examples of the same emerging category: AI models strong enough to raise national security concerns, not only for America but all governing and industry-essential institutions and organizations that depend upon access to these tools to fortify their cybersecurity postures. Recognizing this category, and knowing it now spans more than one developer (and that the leading developers in this space are still US-based), is useful context for any office encountering AI-related cybersecurity questions in the months ahead.
SKILLS HUB
Avoiding LLM lock-in
Chiefs of staff across the Hill are formalizing AI policies and officially designating which institution-approved LLM tool(s) their team is committing to using. For some staff, the office’s endorsement of a tool provides the green light to begin learning this technology and exploring use cases. For others who may be more familiar with LLMs and have a preferred platform, the selection of one tool may lead to frustrations if it is not their personal tool of choice. Beyond initial adoption, chiefs may also have anxiety about choosing one tool at this time knowing that the market may change and the office may shift to use a different LLM in the future.
Individuals and offices can avoid LLM lock-in by building strong foundations for LLM use and remembering that many LLM functionalities are mirrored across all the institutionally-approved models. Here are some tips to ensure your team benefits from their investment in LLM tool creation and use, even if changing platforms:
If your team has created a custom GPT, Project, Gem, or Agent, you can ask that tool to write a memo to be used by another LLM to create a clone of itself.
Sample Prompt: I would like to recreate this project on a different LLM platform. Write me a thorough memo to be read by an LLM that includes all the information it would need to successfully recreate this project. Include a cover page of the memo that includes instructions for me about what content in the memo to paste into the instructions box and other steps I can take to ensure that this project is successfully duplicated.If you have been using an LLM platform and plan to transition to a different one, you can ask the LLM to write you an exit memo for the new LLM to inform it about your preferences, background, and additional helpful context to streamline the transition. Once the LLM has created the memo, review it carefully and update it to reflect your true preferences, and include any additional information you believe would be helpful for the new LLM.
Sample Prompt: I need you to write me an exit memo that I can give to a new LLM account to assist it in getting to know me quickly. The audience for the memo will be an LLM. Please have it include any of my preferences, details about me and the work that I do, and additional context that you believe will help the new LLM be set up for success when working with me. Have the memo include information about the main types of tasks that I’ve gotten your assistance with and any of your lessons learned in working with me.
AI USE INSPIRATION
House and Senate Rules tutor
House and Senate Rules are integral parts of Hill staffers’ knowledge but can be intimidating to learn. Generative AI technologies are being used across academia to create study aid tools for students. The same use case can be adopted by junior Congressional staff and interns working to get smart on their chamber’s rules — fast.
On an LLM that your chamber and office have approved, start a chat with the following prompt and attach the House or Senate Rules.
Sample Prompt: I’m a Congressional staffer with the goal of better understanding the Rules of the [House/Senate]. Attached is the rules package for this Congress. Please act as my tutor to help me learn about the rules in an approachable and memorable manner.
If you would like to create a continual tutoring resource, explore setting up a ChatGPT Project, Claude Project, Gemini Gem, or Copilot Agent that is prompted to be an educational tutor for you.
CAUGHT OUR EYE
Setting expectations: a take on AI’s potential to fix democracy
Legislatures around the world are exploring how AI tools can improve constituent engagement. This piece in Noema by HennyGe Wichers caught our attention for asking a deeper question about what that improvement strives to achieve. It argues that AI tools built to help institutions “listen” to constituents, including Google DeepMind’s Habermas Machine, may be solving the wrong problem if finding consensus is the end goal: democracy was never designed to produce agreement, but to manage disagreement that has no final resolution.
Data Skills for Congress
Applications are now open for the Summer 2026 cohort of the Data Skills for Congress online certification program, designed specifically for Congressional staff offered through UC Berkeley’s Goldman School of Public Policy. RSVP to join the virtual information session taking place on July 8.
Ranked: Which AI companies Congressional staff trust
Capitol CNCT released polling data last week on which AI companies Congressional staff trust, including insights into how those trends in trust align (or misalign) with the LLM tools being used on the Hill.
A case study of using modern technologies to capture institutional knowledge
On July 7 at 11 AM ET, POPVOX Foundation’s Anne Meeker and TheirStory founder Zack Ellis will discuss how Departure Dialogues, a 2025 nonpartisan project by POPVOX Foundation, Niskanen Center, Civil Service Strong, Partnership for Public Service, and Foundation for American Innovation, collected structured testimony from departing federal workers to capture institutional knowledge and elevate insights for potential reforms. RSVP.
CRS and AI
On June 25, the Committee on House Administration held a hearing on the Congressional Research Service and lessons learned from their initial pilots integrating AI into legislative analysis. Key takeaways can be found here. Following the hearing, POPVOX Foundation Cofounder and Executive Director Marci Harris proposed follow-up oversight questions Members should consider, offering a roadmap for how Congress can continue to initiate informative assessments of AI adoption within the Legislative branch. And just a few days later (for $50 + the cost of an Anthropic Pro subscription), Learning Journey AI’s CEO, Nicholas Wagner, released a benchmark prototype and shared the methodology, prompts, results, and code he created. Check it out to compare all major models + CRS on summaries of bills in the 118th and 119th Congresses.
How AI models perform when asked to calculate your taxes
It is no surprise that some Americans turned to AI to assist with navigating their taxes this year. PolicyBench tested the range of publicly available LLMs to calculate the taxes and benefits of 100 households.
@Claude on Slack
For House offices that use Slack, Anthropic has announced a new plugin for Claude, streamlining integration between this institution-approved LLM and the popular workflow management platform.
Using AI to unlock every law in America
In a new academic paper, researchers used AI to find and process every law across localities, counties, states, and the federal government to combine them into one data set dubbed LOCUS: A Local Ordinance Corpus for the United States.
Midjourney goes medical
Midjourney, a popular AI image generation tool, announced the creation of a prototype medical-focused full-body scanner, proposing a future replacement for MRI technologies. AI’s ability to process massive datasets is what makes scanners like this possible.
Anthropic announces release of Claude Sonnet 5
As leading AI companies (such as OpenAI, Anthropic, and Google) continually release new versions of their LLMs, keeping track can be a challenge. On June 30, Anthropic released its newest version of the Sonnet model, Sonnet 5. In the May edition of Keeping Pace, we discuss why model versions matter, and how intentionally choosing which model of an LLM you use can improve the efficiency and quality of your interactions.
Following current events in 1776
As Independence Day approaches, the X account 1776 Live is tweeting news as if we were living in 1776 on the brink of the signing of the Declaration of Independence.
About POPVOX Foundation
POPVOX Foundation is a nonpartisan nonprofit that helps democratic institutions keep pace with a rapidly changing world. Through publications, events, prototypes and technical assistance, the organization helps public servants and elected officials better serve their constituents and make better policy.
