
Note: this newsletter’s open rate, normally a healthy 50-60%, dropped precipitously last week. Either you weren’t interested in hearing me yell at Alex Karp again, or Beehiv’s move to its new email sending platform Hex wasn’t as seamless as they promised. We’ll see what happens this week.
Get ready for a lot of transparency content
For the past few months, I have wandered away from promoting democracy in our online lives and written about other topics: my summer field trip on Google’s corporate culture, AI doomerism, engineer licensing, squirrelburgers, etc. It’s time to lock back in. You have given me positive feedback on my proposals for practical solutions to problems like due process and exit. So next up, I am going to try to do the same thing for transparency.
I have already written a lot about transparency in this newsletter. Transparency is the foundation of any prospective democratic system. We, the people, cannot consent to a form of government that would hide its own workings from us.
Every thing secret degenerates, even the administration of justice; nothing is safe that does not show how it can bear discussion and publicity.
What does “Transparency” even mean?
The ideal of “transparency” has been invoked by online activists to promote a dizzying range of policies. [Just look at the three bills I am watching in the new US Congress, which span everything from revealing popular social media content to auditing the internals of AI foundation models.]
The best framework I have found to organize all of these ideas is Operational Reporting By Online Services: A Proposed Framework, a 2023 piece by Samantha Lai, Naomi Shiffman, and Alicia Wanless.
All three co-authors have impressive experience. For example, Shiffman alone clearly knows more about transparency better than I ever will, due to her work with CrowdTangle (a pioneering tool for understanding viral content that was purchased and eventually shut down by Meta) and as former Head of Data and Implementation for the Meta Oversight Board. Shiffman also co-wrote a great report called The Case For Transparency, which includes over a dozen case studies where social media platform data access led to real-world change.
Anyway, enough fanboying. Here is their paper’s taxonomy:
User level: aggregated information about different types of users.
Platform level: platform architecture and how platforms work.
Policy development and enforcement: internal policies that govern activity and use of the platform (e.g., content moderation policies).
Internal research: the types and findings of research conducted inside the company to understand impacts on users and of interventions.
External requests for intervention: requests to platforms from a third party to act on a user, an account, or a piece of content (e.g., user-flagged content, content removal requests, account suspension requests).
Data access requests and tooling: requests from third parties for access to the personal information of a user or group of users (e.g., law enforcement requests, court orders) and the tools created to facilitate access to data.
Terms of service and privacy policies: terms of service refer to agreements between companies and users regulating the use of the service. A privacy policy is a statement that discloses how and when a company collects, uses, and shares user information. This information is typically shared with third-party services.
Third-party relationships: arrangements between companies and third-party organizations regularly accessing user data.
My take on the framework
There are a few areas of transparency that I’d like to see which aren’t covered in this framework, so here is an iteration. This is just a sketch, compared to the much more thorough descriptions in the original paper. Please let me know what you think at [email protected].
User level: the original paper is great on this.
Platform level: The original paper’s category focuses on algorithmic transparency, but in recent years the potential scope here has expanded dramatically. Meta settled with twenty-nine states over accusations of product design that led to teen addiction, and submitted to an independent auditor of prospective new features. The risks of the AI area have also led to calls for extensive safety testing of foundation models, such as Ro Khanna’s proposal for a new federal agency.
Rules and rule-setting: This category constitutes the quasi-legislative/legal framework that we are required to accept in order to use Internet platforms. It combines half of the original #3 (policy development) and all of original #7 (terms of service and privacy policies). This is also an area of creeping authoritarianism as platforms change the rules to favor themslves (see my piece New data shows platforms use terms of service to subvert the rule of law from March).
Justice systems: The original framework cites enforcement in #3, but law enforcement is just one part of a functional justice system, alongside courts, juries, the penal system, etc. This could also include the original #4, 5, and 6 to cover the role of outside parties like governments, and measurements of effectiveness and fairness.
Safety metrics: Many current debates about social media and AI are arguments about these outcomes. What is the real prevalence and impact of child endangerment, radicalization, harassment, scams, and so forth? This is both about original #4 (internal research) and calls for giving researchers more data access like in PATA.
Data protection: Regulations like the EU’s GDPR require disclosure and consent about how user data will be used, and many jurisdictions require disclosure of data leaks. This could go further to require security audits. This also incorporate the original #8, third-party relationships.
Content: Helping researchers understand what content is popular and viral on a platform is what CrowdTangle did and what PATA would facilitate. This should, critically, consider paid content like ads, not just user-generated content. This could also include proposed disclosures on individual pieces of content, like whether AI was used or if it an advertiser paid to be mentioned. [This is arguably in the paper’s original #1 on user-level data but I thought it was worth calling out separately.]
Financials: We could ask for more regular and specific data on safety and governance expenditures. Today, platforms give out one-time data points that flatter their efforts, like a number of people working on safety or the number of pieces of content taken down, without context or updates. Or, they make implicit financial statements, like a Roblox executive saying last year “there's always more we can be doing” – OK, then why aren’t you doing more? What percent of revenue/profit is going back into safety?
Practical transparency: start with #3, rules and rule-setting
These lists are inspiring, but overwhelming. Maybe the largest tech platforms could manage to comply with all of this at once, but small and mid-sized companies would probably grind to a halt, or just fail to comply and be crushed under fines and penalties. We need to know more, but a transparency regime that just increases market concentration does not seem like a desirable outcome.
Another challenge in many of these categories is that there is no objective standard for how much data disclosure is enough. We haven’t resolved critical questions like aggregate data (totals and summaries) versus census-level; sensitive information about security risks that could be abused by adversaries. And given how little we trust these platforms today, no matter where they draw the line on practical disclosure, we will suspect that the really bad, incriminating data is just on the other side.
So, as a first principle of practical transparency, I am not trying to satisfy all stakeholders. I am just trying to find something that would specifically help advance online democracy, and could be required by governments with an objective standard that platforms could realistically meet.
For me, that means category #3, rules and rule-setting. As you hopefully know by now, my core thesis is that internet platforms should be subject to the laws of the communities they serve. A good first step toward that goal would be ensuring we have absolute clarity around the laws platforms expect us to obey today.
Understanding the rules better will naturally lead us to reasonable questions about how they were made. What right do we have to know who has authority and responsibility for rule-setting? What kind of notice we should get before they change?
A third question this would raise is how well platforms live up to the democratic ideal of equality before the law. We know from revelations like Meta’s Cross-Check program that platforms have special procedures in place for high-profile accounts like politicians and popular creators. Should platforms disclose these programs? Should they even exist at all?
In coming weeks, I will dig into each of these questions, in search of feasible transparency improvements that could help platforms re-earn our trust. Stay tuned.
Ideas? Feedback? Criticism? I want to hear it, because I am sure that I am going to get a lot of things wrong along the way. I will share what I learn with the community as we go. Reach out any time at [email protected].

