Meta-Layer Initiative kickoff

Daveed Benjamin opened the Zoom with Liz and Brad as co-organizers. The group was asked to stay muted unless speaking, raise a hand or use chat for questions, and treat the session as a working meeting rather than a lecture. Note-takers were already in place, and Liz planned short breakouts after the opening remarks.

Daveed introduced Vint Cerf as one of the fathers of the Internet, Google's Chief Internet Evangelist, and a reader of The Metaweb who had already commented on the book. Cerf joked that he had words, but whether they were wise was for others to decide.

Cerf framed the proposal as infrastructure layered on top of the existing Internet–or whatever the Internet becomes. The lesson from IP, he said, is to stay agnostic about the layer below: IP does not care whether bits travel over radio, fiber, or copper. It only asks whether a bundle of bits can be delivered with some probability greater than zero. Recovery, reliability, and low latency live above that.

He extracted several design implications from the Internet's history:

• Standards. Interoperability cannot depend on pairwise negotiation. Standards let parties who will never meet still interwork, and they let competing products share properties so customers have choice.
• Broad adoption. Infrastructure only pays off if it is widely used.
• Self-sustaining business models. Early Internet backbones were government-sponsored (DARPA, NSF, DOE, NASA) and limited by acceptable-use policy. Cerf described connecting MCI Mail to the NSFNET backbone in 1988 as a deliberate break of that policy. Commercial ISPs (UUNET, PSINet, CERFnet) and equipment vendors followed. Some firms failed or were acquired; the point was that a business model had to support both spread and ongoing enhancement.
• Simplicity. Prefer one way to do a thing. Multiple options force a negotiation before two parties can talk–the problem he associated with X.25 and OSI's many transport protocols.
• Safety, security, and strong authenticity. Because the layer will touch identity and high-consequence transactions (finance, medicine), it must resist attack. AI and large language models raise the stakes: they will be invited to act on users' behalf, and they hallucinate. His elevator example: once the door closes, you depend entirely on someone else's software.

Demographics, he added, will push societies toward robotics and AI taking actions people once took. Humans already make mistakes; software will too. The design question is how to reduce the likelihood of costly ones. The most important question before implementation: what could possibly go wrong?

Accountability and agency. Daveed asked about accountability. Cerf said his two watchwords are accountability and agency. He once favored anonymity as a default. He still believes people should not have to identify themselves merely to use Internet resources–but online systems now amplify everything, including harm. Cross-border damage is hard to police (he noted the UN cybercrime treaty debate and the Budapest Convention). Accountability requires identification. Agency is the other side: strong authentication protects you from software or people taking unauthorized actions in your name. Any agent acting for a person needs carefully designed transitive authority–no implied consent.

Why pricing was left out of the protocol. Eric Harris-Braun (Holochain) asked who decided not to put cost and pricing into the Internet's protocol design. Cerf said it was circumstance, not a commercial theory: ARPANET and the Internet were paid for by DARPA as command-and-control infrastructure, not as a market. Kahn and Cerf published the 1974 IEEE design so allies could implement it–and they published it openly because they could not know who the allies would be 25 years later.

Encryption after quantum computers. Asked which encryption standards would be needed, Cerf said RSA-style algorithms that rely on factoring are under a future quantum threat (Shor's algorithm). He did not believe a quantum computer could yet break significant crypto, but information that must stay secret for decades needs post-quantum algorithms now. NIST had collected dozens of lattice-based candidates and down-selected a small set for signatures and encryption. Google had begun implementing them.

What might be next. Quantum entanglement, he said, does not give zero-delay communication. If there is a next big thing, it is likely AI. LLM failures, in his view, are failures of context: a chatbot asked to write his obituary invented a death date, conflated other people's careers with his, and fabricated family members. Larger context windows help generation but do not fix training that lacks adequate context. He pointed toward smaller specialist models collaborating–closer to Minsky's Society of Mind–before trusting these systems to act in the world.

After Cerf's remarks, Daveed stated the meeting's purpose: launch a collaborative of people and organizations willing to help build an application substrate for an environment above the webpage–what he had been calling the Overweb. The aim was a common identity space and security model, plus a way to manage how AIs interact in that environment, kept as lightweight, protocol-agnostic, and simple as possible.

The inspiration was context. Around 2015 he saw the commercial web starve outbound links: social-media feeds replaced web surfing, and platforms were rewarded for keeping people on-site. The work, he said, is also older than the Web. Annotation is a thousand-year human practice that was present in early knowledge systems and in Mosaic, then stripped from the browser under commercial pressure. Marc Andreessen later invested in Rap Genius in part to restore layering knowledge on knowledge–and wondered what the world would look like if annotation had stayed in the browser.

Liz facilitated roughly 10–12 minutes of breakouts (one round, not five-minute hops) around three questions. Groups could spend unequal time; they were asked to bring back at least one useful answer:

1. What unique, valuable contribution can this group make to the Meta-Layer–not everything that could be done, but what this constellation is best positioned to do?
2. What would a meaningful first milestone look like?
3. How should the group organize itself to reach that milestone?

In the host/note-taker room, Cerf argued against boiling the ocean. A focused contribution could be a standard for federated strong authentication–multiple competitive authenticators, with oversight analogous to how browsers treat certificate authorities. He then named a concrete exercise: an essay titled Desirable Properties of the Meta-Layer. Making properties specific would reveal the ensemble of things the layer must implement. For any chosen property, ask what would have to be true, what assumptions and agreements are required, and who is accountable if an authenticated party behaves badly.

That room also discussed a cultural shift: more credibility for people who deliberately identify themselves; provenance for information (as in the art world); nanopublications as a way to open micro-scale scholarly contribution; and the downside of open feedback (false copyright claims against YouTube). Strong authentication creates a path to reputation, compensation, and “penalty box” feedback loops. Badging, a landscape of existing overlay projects (Hypothesis and others), and a glossary–not only a word cloud–were suggested as supporting artifacts.

Liz asked each room for a short readout. Themes that surfaced:

• Identity, privacy, and proof of humanity. Annotation needs trust in who is writing, and many annotators still need privacy or anonymity.
• Commerce and safety. How the layer might support exchange without becoming an unsafe space.
• Standards from implementation. Interoperability and open process; standards that emerge from communities actually building, then coming together.
• Use cases still unclear. Several people said they did not yet see what would motivate ordinary users.
• Adjacent work. The liminal web, Goodly Labs / Public Editor (annotating news for manipulative language), and other protocols that might link these efforts.
• Shared target and funding. Many disciplines and some corporate walls; a common outcome, a roadmap, and a clear funding path (new foundation vs. existing apparatus) were named as organizing needs.

Daveed reported the host room's proposal: focus on a standard for federated strong authentication; write an essay on the desirable properties of the Meta-Layer as the first milestone; map who is already operating in this space; and synthesize qualitative and quantitative notes from the collaborative document.

The formal meeting closed with thanks to Cerf and the participants. A smaller group stayed on. Brad asked for visuals and simple UX experiments so people could picture the idea. Others tied that to funding: a first milestone should be something you can put in a grant proposal; short-, mid-, and long-term goals; and business models, not only vision.

On use cases, Daveed said they are as vast as the Internet itself–more context on any page. Liz pushed toward the art of the practical: one pinprick human problem, solved with low friction, rather than another decentralized solution searching for a problem. Trust was named as a pillar, not a late add-on–trust in outcomes, in operators of infrastructure, and in authentication as the basis for accountable collaboration.

Mental models that landed: a community “data water main” that clients can tap; start with a specific community (as the early Internet started with a defense community and its allies) and reduce friction between communities; start with trust, not end with it; and meta-community–knowing that people you meet above a page are real people in good standing, with presence and artifacts that persist across the Web.

A late technical thread covered ledgers, browser tech, Web3, and AI interfaces. Cerf had already pointed to NIST post-quantum work. Stay-on participants warned that new algorithms are easy to implement badly (Heartbleed as a caution), hybrid deployments may be safer than a sudden cutover, and the group should keep anonymous, private, secret, and confidential from collapsing into one word.

Commitments from the room that day:

• Email everyone who attended and those who could not, with a link to the collaborative notes.
• Ask participants to add breakout insights and later reflections to that document.
• Propose a first milestone to the group and invite people to organize around it. The leading candidate was an essay on the desirable properties of the Meta-Layer, with federated strong authentication as an early property to make concrete.
• Draft a landscape of projects already operating in this overlay space, plus visuals that make the user experience graspable.
• Treat funding and a practical, trusted first use case as part of the organizing work, not a later afterthought.

That milestone is the work now underway. The essay became the Desirable Properties drafts, the Challenge, and the book in public review.