Piero Bosio Social Web Site Personale

By Ivan Sigal, Jessica Theodule, Joshua Tan, Sarah Nicole, and Mallory Knodel.

Open social protocols promise to deliver user agency. As they become more widely adopted, users experience different and new tradeoffs in the technical, economic, and governance aspects of open, federated architectures such as DSNP, ActivityPub, and AT Protocol, in comparison to the walled-garden platforms of a handful of big tech companies. The top-down control of dominant social media platforms manifests in centralized data accumulation, opaque ranking systems, surveillance tech revenue models, and locked-down architectures. These practices in turn contribute to widespread violations of user data protection and privacy, which are also associated with the algorithmic feeds and large language models that power big tech companies’ AI products.

Reconfiguring agency in the age of agentic web 

User agency, or the ability of people to control their digital environments and make choices in their own interests, is facing forced reconfiguration in the emerging agentic web. Agentic AI's growing presence and possible dominance is being driven by massively capitalized companies that organize economic power, technological systems control, political and regulatory influence and social persuasion. They are setting the terms for information futures and embedding the rules for agentic AI into every aspect of socio-technical architectures. This is happening in the breach of regulatory controls and oversight, and without substantial public input or debate into decisionmaking. 

In this vision of the agentic web, user agency is being remade: the very existence of functional user agency is not guaranteed. The many challenges to the agency of users include the ability to retain control over their data, their privacy, and the organization of their content sources. Perhaps most importantly, users' ability to meaningfully contribute to the building of the open web is itself a field of contestation. This contestation intensifies as AI agents begin to occupy roles once reserved for human participants, from curating and posting content to even acting as proxies within the very social infrastructures that user agency was once understood to have built. 

Open social protocols offer alternative visions that, in a range of ways, can contribute to a renewal and long-term instantiation of user agency. Their openness, meaning the ability of people to use and extend protocols in accordance with open software licences, creates the possibility of  mixing and matching open social protocols with an AI agent-mediated web. AI agents may indeed become content creators and curators, but people may also retain the ability to manage their relationships to the output of those agents. 

Compellingly, the open social web itself may be a source of data to build public AI models. These models offer alternative approaches to the brute force, totalizing and extractive commercial models that currently dominate the landscape. In the long term, open approaches can succeed because they allow many contributors to combine, remix, and extend models in decentralized forms, just as open architectures beat closed systems in an earlier generation of web technologies. A dialogue between open social tech and public AI begins with values, ethics and a common stance toward governance, and extends through protocols, development strategies and alternative models of organization, from governmental models to non-extractive commercialization, to collectives and cooperatives, to community-driven nonprofit initiatives. This future is also not guaranteed, but a growing number of technologists, ethicists and creators are eager to participate in its manifestation. 

The technical, economic, and governance choices shaping the emerging agentic social web will determine whether AI reinforces centralized platform power or enables more distributed, accountable forms of participation. Critical conditions include standards and governance mechanisms that protect data sovereignty, support accountable AI agents and sustainable business models, and avoid replicating the extractive dynamics of old social platforms. 

Below we unpack what we mean by “open social web” and “public AI”. We examine the current trends and motivations in building both, and how they might be built together, with intention and according to principles that lead users away from top-down, walled-garden platforms and towards a common means of production and consumption that is truly democratic.

The Social Web from identity, federated architecture, to open governance

The social web encompasses technologies and processes that support social interaction online. Conceptually this includes social media platforms that focus on social networking and user-generated content, and on the myriad connective protocols and tools that support sociality across the internet. 

Open social web protocols offer a common language to organize information interaction in support of social connection. As protocols support email interoperability regardless of host or service provider, so do open social protocols offer the ability for social technologies to interact as infrastructure. Key areas of interaction include identity and social graph interoperability, data portability, and permissionless component development.

Open social web protocols encode assumptions about power, identity, governance, and enforcement. A variety of emergent protocols offer a plurality of visions for how we might organize decentralized and open social tech.

• ActivityPub (AP) is a W3C standard for federated social interactions. It defines how actors publish and receive activities across a network of servers operated independently. Federation enables plural governance but assumes local, human-scaled moderation, which strains under automated agent traffic. 
• AT Protocol (ATProto) separates identity, data storage, ranking/aggregation layers, and application clients. Data and identity portability are emphasized, enabling users to move both social graphs and data across services.
• Decentralized Social Networking Protocol (DSNP) positions the social graph as user-owned infrastructure rather than platform exhaust. It reframes access governance as collective state rather than vertical platform control, enabling novel approaches to delegated authority and consent. 

Other open social network protocols extend the design space:

• Nostr decouples storage from identity via public-key posting to relays, prioritizing censorship resistance and portability. 
• Secure Scuttlebutt (SSB) offers peer-to-peer network replication without central servers, emphasizing cryptographically verified feeds. 
• OpenAutonomy provides early decentralized interoperability across internet domains without central mediation. 

These protocols illustrate that decentralization is not monolithic: federation, peer replication, cryptographic relays, and user-centric identity are complementary paths with different governance implications.

Personal identity and authentication mechanisms provide the foundational mechanisms for authentication and data control. They determine who controls data and under what terms it is shared. 

AI plays an increasingly important role across the entirety of the social web. It is being embedded at every stage of network function, from coding support and automated work flows, to search and discovery, content moderation, to user customization. The ethics of AI use cut across each of these aspects. Legibility of AI inputs, governance and regulation, and intellectual property and fair use are all zones of contestation.

In content space, bots and agents jostle with humans for attention and information creation. Narrative formation, information operations and manipulation, deepfakes, privacy, and user control over the degree of interaction with AI agents are all unsettled across the social web, including in open social spaces.

The Agentic Web: The Architecture of Agency

Public AI models and infrastructure hope to offer alternatives and reduce dependence on extractive commercial AI. Large language models, previously experienced through chatbots like ChatGPT or Claude, are increasingly being fitted into more autonomous planning and tool-using systems called “agentic AI”. While the language of agents implies autonomy and some form of delegated authority, most agents are not yet acting independently on behalf of users, wielding email addresses, social network accounts, or credit cards. In practice, today’s agentic AI focuses more on multi-step planning workflows (e.g. sequences of LLM and tool calls) while tool-use tends to focus on search of the web and file systems. However, coding agents with limited autonomy have become explosively popular, and the recent release of the open source OpenClaw agent framework presages a more autonomous future for AI agents posting on messaging platforms and organizing on social networks.

These possibilities create a new set of requirements for identity, capability, discovery, trust, governance, and interoperability. They also demand more investment in public standards, protocols, and infrastructure. Below, we review a number of protocols operating in this space.

A number of core protocols already structure agentic behavior, especially online.

• Model Context Protocol (MCP), piloted by Anthropic, provides a technical layer to standardize how models receive structured context from external sources. It enables tools and services to communicate context to agents, but does not by itself resolve governance issues about who authorizes, controls, or adjudicates that context.
• Agent-to-Agent (A2A) Protocol, piloted by Google and Linux Foundation, enables agents to find and communicate with each other using standardized mechanisms.
• Agent Skills, also piloted by Anthropic, are an open modular format encoding procedural knowledge and workflows that agents can discover and load on demand.

Other agentic protocols are still emerging or proposed. Many of these are reacting to the need for private, contextual data like project data and history, preferences, workflows, relational history, and personal metadata. These protocols differ in scope and maturity, but collectively they point toward a layered agent infrastructure that supports decentralized, interoperable, and accountable, multi-agent systems.

• OpenClaw, just recently acquired by OpenAI, is a popular framework that orchestrates agentic access to a range of messaging, email, and workspace applications, enabling more assistant-style workflows.
• Human Context Protocol (HCP), is a theorized framework that reframes delegation as a governance layer that specifies non-goals, scope, revocation, and norm agreements that agents must adhere to when acting for users.
• Google’s new WebMCP and startup Unternet’s Web Applets are both protocols for enabling web actions by agents.
• AgentMail, a Y-Cominator-backed startup implementing an agent-first email client. While not strictly speaking a protocol, it’s important to understand how agent email might evolve because email often functions as an identity primitive.

Personal data control

Agentic systems invert the traditional user-to-platform data flow of social platforms. Instead of users visiting platforms, agents might also collect data in addition to generating it. This creates opportunities for more personal data control on the social web, but also raises acute data sovereignty concerns about where private context is stored, if memory is centralized, can users export or fork their agent history, and how to achieve privacy-preserving training on interaction data.

There are a range of different options for asserting such data control:

• Existing authentication protocols such as the W3C’s decentralized identity (DID) and verifiable credentials standards might bind agent identity to cryptographically verifiable credentials, addressing trust and accountability.
• Personal data might be stored and processed in user-controlled stores (e.g., cryptographically secured servers), and agents access context through permissioned interfaces. Under this model, agents become ephemeral executors rather than data aggregators.
• Protocols and interoperability enable portable agent memory, which is particularly challenging across AI systems today.
• Public AI models and infrastructure could further reduce dependence on extractive commercial options. However, public models, inference, and training data can mitigate but not solve the governance problem of contextual memory. The locus of control over personal metadata will determine whether agentic AI enhances or erodes human autonomy.

Aligning technical and economic incentives

AI systems shape online discourse and their development is dominated by centralized platforms. In this sense, AI systems exploit and potentially amplify the network effects that have long shaped social media's trajectory. Open social web protocols risk inheriting the same inequities if they do not embed human rights, democratic governance, and sustainable economics into the AI layer from the outset. How can we build bridges between open social web protocols and agentic AI so that builders and experts are mutually reinforcing rather than parallel and disconnected?

AI systems increasingly shape online discourse at scale, and when combined with data accumulation, scale becomes a mechanism through which power, visibility, and participation are unevenly distributed. We have intuition about the power wielded by centralized data systems. Will de-centralized data systems be inherently better? What novel risks can we identify?

As AI agents expand across functions such as creation, moderation, and curation, the unchecked broadening of system scope risks complexity beyond the reach of norms, regulations and law. At the same time, hyperlocalization is also more possible and perhaps desirable, such that distinct social contexts represented by a single, governable surface no longer makes sense.

Democratic, multistakeholder governance is meant to hold it all together because without protocol-level governance mechanisms that enable contestation, refusal, and accountability, open social systems risk reproducing platform dominance under new technical forms rather than redistributing power.

Open social web protocols and public AI will not be governed primarily by stated values, but by the incentive structures embedded in their technical and economic design. Strong business models for open social web and public AI have yet to emerge. We now have the opportunity to interrogate existing implementations’ business models, whether they are sustainable, and what are the gaps in fulfilling their ambitions? Then it becomes crucial to anticipate how values might be impacted by profit motives, with an aim to influence outcomes that improve upon the status quo.

We stress the importance of continued dialogue from various perspectives globally and at all layers of this complex technical stack, from users to implementers, in order to create shared problem definitions, surface emerging priorities, and identify concrete next steps across research, standards development, and governance communities.

Want To Appear Here? 💡
Sponsor Internet Exchange 🩷

Support Internet Exchange and reach the people shaping internet governance, standards, and digital policy. We offer a limited number of sponsorship slots for values-aligned organizations launching research, reports, books, or events in public-interest technology. Learn more about sponsorship options and how to book.

This Week's Links

Open Social Web

• Startup Germ Network has integrated its end-to-end encrypted (E2E) messaging service directly into Bluesky, making it the first private messenger that can launch natively from within the Bluesky app. https://techcrunch.com/2026/02/18/a-startup-called-germ-becomes-the-first-private-messenger-that-launches-directly-from-blueskys-app 

Internet Governance

• Leaked documents suggest the US Department of Homeland Security has expanded a program allowing agents to use fake social media identities to join private groups, friend users and gather intelligence. A program that could enable widespread, hidden surveillance of online communities. https://www.kenklippenstein.com/p/exclusive-ice-masks-up-in-more-ways 
  • Related, The US Department of Homeland Security has sent hundreds of administrative subpoenas to tech companies seeking the identities behind social media accounts that track or criticise ICE. https://www.nytimes.com/2026/02/13/technology/dhs-anti-ice-social-media.html 
• During a fireside conversation with Motion Picture Association chairman Charles Rivkin, India’s IT minister Ashwini Vaishnaw said the country is working with more than 30 nations on shared legal and technical measures to tackle AI misuse, including watermarking and labelling of synthetic content. Our take? The involvement of the MPA suggests the conversation may focus narrowly on intellectual property. It should also address broader societal harms. https://timesofindia.indiatimes.com/india/india-in-talks-with-30-nations-to-curb-ai-misuse-ashwini-vaishnaw/amp_articleshow/128438755.cms 
• AI sovereignty depends on open, interoperable standards rather than countries building their own AI stacks from scratch, write Eileen Donahoe and Konstantinos Komaitis. They urge governments to focus on procurement rules, international standards, data portability and cooperation to avoid vendor lock-in and retain control over how AI systems are integrated and governed. https://www.techpolicy.press/why-ai-sovereignty-depends-on-interoperability-standards 
• UK Prime Minister Keir Starmer warns Elon Musk and other tech executives that “no platform gets a free pass” on illegal content in the UK, and moves to tighten online safety laws following xAI’s a deepfake scandal. https://www.ft.com/content/15917aa4-2d40-49be-85c3-da395b16e7f1 

Digital Rights

• A coalition of UN bodies and more than 50 organisations has issued a joint statement calling for a child rights-based approach to AI, urging coordinated global action to ensure AI systems respect, protect and fulfil children’s rights. https://www.itu.int/hub/publication/d-str-cyb_joint-2025 
• A recent wave of bot-driven cyberattacks has targeted the social media accounts of Serbian media outlets, prompting alarm from the Media Freedom Rapid Response partners and ARTICLE 19 Europe. https://www.article19.org/resources/serbia-rising-bot-attacks-on-independent-media-instagram-accounts 
• A Citizen Lab report claims Kenyan authorities used Israeli Cellebrite software to break into activist Boniface Mwangi’s phone while he was under arrest. https://www.theguardian.com/world/2026/feb/17/trim-slugs-to-kenya-authorities-israel-tech-crack-activists-phone-report 

Technology for Society

• Grassroots opposition to AI data centres is rapidly growing across the US, becoming a new focal point for climate activism, anti–Big Tech sentiment, and local democracy. https://newrepublic.com/article/206633/data-centers-ai-big-tech-opposition 
  • Related: A new report argues that claims AI will help solve climate change are overstated and often used to justify the rapid expansion of energy-hungry data centres. It finds most climate benefits come from older, smaller forms of AI, while generative AI is driving rising fossil fuel use and weakening tech companies’ climate commitments. https://ketanjoshi.co/2026/02/17/big-tech-greenwashing-report 
• Despite previously warning about the dangers of autonomous weapons (anyone remember that open letter?) Elon Musk’s SpaceX and xAI are now competing in a Pentagon $100 million prize challenge to build voice-controlled, autonomous drone swarm technology. Thanks, I hate it. https://www.bloomberg.com/news/articles/2026-02-16/spacex-to-compete-in-pentagon-contest-for-autonomous-drone-tech 
  • Meanwhile, Anthropic is seeking limits on the Pentagon’s use of its model for surveillance and fully autonomous weapons, prompting Defense Secretary Pete Hegseth to threaten cutting ties and labelling the company a “supply chain risk.” https://www.axios.com/2026/02/16/anthropic-defense-department-relationship-hegseth 
• A new report from the Association for Progressive Communications warns that the digital divide is widening, with many people still offline or only “barely online” despite global connectivity efforts. https://www.apc.org/en/pubs/tree-known-its-fruit-growing-community-centred-connectivity-inclusion-equality-and 
• Rebecca Ryakitimbo and Risper Arose reflect on a Feminist Tech Exchange workshop exploring feminist mesh networks as community-owned, care-centred alternatives to traditional connectivity, grounded in equity, autonomy and collective power. https://www.apc.org/en/blog/weaving-networks-care-and-creativity-reflections-feminist-mesh-network 
• Migration and Technology Monitor Fellows sharing their work on Open Democracy: 
  • Refugees face widespread, often untreated mental-health trauma. Aqila Abdelkarim Ali writes that her platform, MentalEase, can make support more accessible, private and culturally relevant. https://www.opendemocracy.net/en/author/aqila-abdelkarim-ali-somaliland 
  • Refugees are often reduced to statistics and denied control over their own narratives. Simon Drotti’s, Memory Scroll, gives refugees a secure space to tell and preserve their stories on their own terms. https://www.opendemocracy.net/en/memory-scroll-refugees-tell-stories-social-media-bidibidi-uganda 
  • Migrant workers are increasingly exploited through complex digital systems, fraud and financial misinformation. Rajendra Paudel highlights how digital and financial literacy can help migrants navigate technology, contracts and migration systems with dignity. https://www.opendemocracy.net/en/fake-visas-tech-role-exploiting-migrant-workers 
• Manifesto For Technology As a Tool to Strengthen Civil Society: A manifesto drafted by MTM fellows that can be commented on, adapted,and translated as it travels across communities. https://pad.riseup.net/p/Migration_and_Technology_Monitor_Manifesto-keep 
• This year marks 100 years since the birth of Margot Frank, Anne Frank’s sister. While her diary has been lost to history, photographs, letters, and memories from those who knew her remain. Explore them: https://margot100.annefrank.org 
• A new Tech and Social Cohesion essay brings together four perspectives arguing that AI will amplify either cooperation or division depending on how it is designed, urging developers and policymakers to centre trust, relationships and human flourishing in AI systems. https://techandsocialcohesion.substack.com/p/the-ai-cooperation-test 

Privacy and Security

• CISA is urging highly targeted officials to use end-to-end encrypted messaging, specifically recommending apps like Signal to protect sensitive communications from interception. https://www.cisa.gov/sites/default/files/2025-12/guidance-mobile-communications-best-practices_508c.pdf 
• Meta is planning to add facial recognition to its smart glasses, allowing users to identify people and access information about them through an AI assistant. Internal documents show the company has discussed launching the feature during a politically busy period because it expected civil society groups and critics to be distracted by other political issues, reducing backlash when the feature launches. Why do they put this stuff in writing‽  https://www.nytimes.com/2026/02/13/technology/meta-facial-recognition-smart-glasses.html 
• AI systems that remember users by drawing on emails, searches, photos and conversations create a major new privacy risk. By storing and combining personal data across contexts, AI agents could expose a detailed picture of people’s lives and let sensitive information leak into unrelated decisions. Miranda Bogen and Ruchika Joshi in MIT Technology Review. https://www.technologyreview.com/2026/01/28/1131835/what-ai-remembers-about-you-is-privacys-next-frontier 
  • Related, A New America brief warns that emerging standards like the Model Context Protocol could turn persistent, cross-service “memory” into a new source of privacy, security and competition risks without early governance. https://www.newamerica.org/oti/briefs/ai-agents-and-memory  
• The Global Cyber Alliance’s 2025 annual report outlines how the organization is helping individuals, nonprofits, and infrastructure operators respond to rising AI-driven cyber threats through training, tools, and global collaboration. https://globalcyberalliance.org/report-2025-gca-annual-report 

Upcoming Events

• Building a New Social Web: Protocols, People and the Promise of a Democratized AI-Data Economy. Mallory is travelling to India to launch the first event in a global series focused on AI and the social web, held alongside the AI Impact Summit in Delhi and hosted by the Observer Research Foundation. February 20, New Delhi, India. https://socialwebfoundation.org/2026/02/06/ai-and-the-new-social 
• PDN Pro-Social. Building Better Feeds. Jonathan Stray explores promising algorithmic approaches to reducing dangerous division. February 20. Online. https://luma.com/n8649ikc?tk=TvT0LG
• A new webinar introduces the Trustable AI Bill of Materials (TAIBOM), a framework designed to document and audit AI systems across their full lifecycle to improve transparency, accountability, and security. February 23, Online. https://aistandardshub.org/events/building-the-trustable-ai-bill-of-materials-taibom 
• Journalism Technology London Meetup. February 24. London, UK. https://luma.com/8s7g1fdm 
• Fourth Global Symposium on Technology-facilitated Gender-based Violence (TFGBV). March 4-5. Remote. https://tfgbvglobalsymposium.vfairs.com 
• Palestine Digital Activism Forum (PDAF) 2026 hosted by 7amleh – The Arab Center for the Advancement of Social Media. An amazing speaker lineup!  March 30-31. Online. https://events.ringcentral.com/events/pdaf-2026 
• Mobilise supporters & build campaign momentum. A one-day online accelerator for NGOs stuck on how to mobilise people, build momentum, and turn ideas into action. April 15, Online. https://www.resource-alliance.org/event/creative-organising-lab 
• Take Back Tech 3 is a gathering for organizers, artists, tech workers, academics, lawyers, and more to rally together and strategize our next power-building moves. April 17-19, Atlanta, GA. https://www.takebacktech.com

Careers and Funding Opportunities

• Center for Democracy & Technology: Senior Technologist, AI Governance Lab. Washington DC, Hybrid. https://cdt.org/careers/#op-690036-senior-technologist-ai-governance-lab 
• AI-Human Research Center at Carnegie Mellon University: Post Doctoral Fellow — Generative AI, Wellbeing, and Customer Support. Pittsburgh, PA https://apply.interfolio.com/180675 
• Digital Investigations Lab: Solutions Technologist. Multiple Locations. https://job-boards.greenhouse.io/humanrightswatch/jobs/8409308002 
• The International Panel on the Information Environment (IPIE): Director of Policy and Strategy. Remote. Proximity to London or Zurich is advantageous. https://www.ipie.info/careers/director-of-policy-and-strategy
• Global Partners Digital: Policy & Advocacy Officer. Remote UK. https://apply.workable.com/global-partners-digital/j/B134548129/?Join 

Opportunities to Get Involved

• Governments around the world are using firewalls to block access to information, directly infringing on the Universal Declaration of Human Rights. You can help by using your unused domain to redirect visitors to uncensored news, Wikipedia and other blocked resources. https://domainsforaccess.com 

What did we miss? Please send us a reply or write to editor@exchangepoint.tech.

Curious about human rights and internet and web standards? Check out @mallory@techpolicy.social on the revolution.social podcast with @rabble.nz@bsky.brid.gy here or wherever you get your podcasts!
Youtube Video