Michelle Lee (Mosh) gives a talk called Beyond the Skeet ATProto's Potential for Digital Preservation and Human Rights at ATmosphere Conference Seattle 2025

Great, thank you. Hi everyone, it's great to be here this weekend. Really fun vibes here at this event, so thank you so much to all the organizers, speakers, and volunteers. I'm Michelle Lee from the IPFS Foundation. IPFS is a project that spun out of Protocol Labs last year after about 10 years. We have been building building blocks for working with content addressing and decentralized networks.

I've been working in tech for a long time, mostly at the intersection of software systems and open society. For my first job out of college, I interviewed with one of these social media platforms. I didn’t take a second interview because I thought, “Oh, I want to work on something more important than silly little online blurbs.” I don’t really have any regrets, but wow, I was very wrong about the importance of silly little blurbs, because after 10–20 years, they have changed from a toy that rand on the internet was playing with to a real global force.

Is social media a good thing or a bad thing for democracy? Pew Research showed in 2020 that most people said social media is a good thing for democracy. The US is an outlier—if you look at the graph, the US is the second one down, skewing very negative. But almost all the other respondents still believe that social media is good for democracy. So, speaking of hope, there you go.

There isn’t a separate imaginary world online—people share things from their real-world lives: thoughts, feelings, photos, events, organizing—bringing our real-world life into the Matrix and then using what happens online—the discourse, conversations, calls to action—to change minds or create movements offline.

So other than AI jokes, what makes up a movement? There are a lot of different things: policy pressure, community building, educational outreach, rapid mobilization. A lot of these have been in full view over the past 15 years. I want to take a quick deep dive into the last two: eyewitness reporting and documentary evidence as used on social media.

I met a group from the Sudan Human Rights Network last month. They coordinate more than 40 grassroots groups in Sudan. They maintain archives, investigate and verify incidents, and coordinate resources. They have a few different archiving projects or focuses, for both open source and closed source media—open source being publicly accessible social media content, and closed source being direct submissions of video or photos from their partner organizations.

These examples are submitted and reviewed by humans before being added to an archive or shared with their partners for near-term action like monitoring or aid. Eventually, this evidence may be submitted to the International Criminal Court. The key takeaway: submitting a link to a social media post is just one of at least eight steps in how organizations like these use, reuse, and make sense of what happens online. How a post is packaged, the metadata included, and how easy API access is—all of that matters to how useful and impactful a social media post can be and how it lives on after the initial event.

Another example is the Starling Lab, a collaboration with the USC Shoah Foundation and Stanford University to build a framework for journalistic integrity. This includes tools to hash images with IPFS CIDs, inspect the source and provenance of photos, and even partner with phone and camera manufacturers to bake EXIF data like GPS location and timestamps into the image metadata.

I know someone this morning suggested scrubbing all metadata from images. I just want to urge you to consider that—sometimes anonymity is the goal, but sometimes provenance and verification is the goal. Starling published an interactive article in Rolling Stone last year—if you click the button in the lower left of the photos, you can inspect it with Starling tools and see all the metadata. Blue Sky also has the same verification built in through IPLD, which makes it a really powerful primitive—not just for designing open networks and growing a community of 33 million users, but also for what you can do with this media outside the network.

The key takeaway: social media artifacts are part of information journeys and pipelines far beyond what fits in a text box, and we should be building accordingly.

I want to seed a few ideas for App Protocol and our community to make a better social media and web landscape.

• Longevity: App Protocol has been designed to outlast Blue Sky the company. Blue Sky is an impressive alternative to Twitter, but in middle-income countries, most people don’t use Twitter. Only 10–25% are on Twitter, but 60–90% are on WhatsApp and Facebook. These platforms prioritize peer-to-peer and interpersonal communications over broadcasting, and these are forms of social media that are incredibly powerful and influential.

• Policy: Standardizing user intent and policies globally would help pipelines like the ones I mentioned earlier. Blue Sky is one platform among many; standardizing intents or policies for data usage makes it easier to incorporate them into platform-wide tools.

• Fighting bots and misinformation: Posts on social media can solidify sentiment within 50–70 minutes. If something is going to go viral, it happens basically within an hour. That’s your small window to fight misinformation before it spreads.

• Content creation: When video streams, images, or photos are captured, processed, and verified, preserve the metadata and content address everything. This becomes valuable for moderation, reuse, archiving, and processing.

• Export and archiving: Integrate with existing web archiving tools like Web Recorder.

• Analysis tools: Build tools that incorporate events outside the social network—weather, natural disasters, election cycles, album releases—and correlate them with activity in social networks.

Within the IPFS project, we have two major technical and product focuses this year:

• Decentralization: Moving away from reliance on hosted gateways.

• Modularity: Making IPFS components more practical to use independently.

We have three initiatives:

• Dazzle: A strict subset of IPFS, CIDs, and IPLD—lightweight and easy to use. Published a spec in December, nine implementations today. Part of a family of pun-named specs: Dazzle, Razzle, Masle, Bazzle—designed to make content addressing easier to use in web applications. Preparing for an IETF submission in July.

• CID Profiles: Ensuring deterministic CIDs with a standardized system of profiles. Currently, defaults are complex and arcane; profiles make it easier to know what you’re asking for. Proposal discussion on the IPFS Dev forum in April.

• CID Congress: Online meetings for content addressing enthusiasts to share work, ideas, and improvements on CIDs, IPLD, CAR files, etc. Next meetings in April and June.

Finally, we’re running a grants program focused on data utilities that work with IPLD, IPFS, CIDs, etc. Small to midsize grants in three focus areas: CAR archive tools, Dazzle testing suite, and improving data utilities for content addressing.

Thank you to the teams and people who’ve been building content addressing tools over many years—Starling Lab, Numbers Protocol, the Sudan Human Rights Hub, and everyone shaping these projects and initiatives. You can find me on email, Blue Sky, or in person here. Thank you very much, and keep building in the free world.

The videos from ATmosphereConf 2025 held in Seattle, Washington, are being republished along with transcripts as part of the process of preparing for ATmosphereConf 2026, taking place March 26th - 29th in Vancouver, Canada.

Follow the conference updates and register to join us!