16 key talks from Devcon 8 in Mumbai and the official news from the week.
Vitalik Buterin · 11.03
He diagnosed that AI is shaking existing assumptions about software security, device privacy and personal data. Technology monopolized by a few is becoming the default path, so he argued that defensive decentralized acceleration (d/acc), which defends with open and verifiable technology, is needed.
Vitalik Buterin compared the past year, as reshaped by AI, to a World of Warcraft expansion pack. Everyone's level rose at the same time, so monsters that used to be hard became easy, but the gear and skills people had built up were reset and new threats appeared. He also pointed out that we have entered an era in which software bugs are exposed one after another and a TV can spy on its users.
Returning to the d/acc idea he published in 2023, he named three extremes facing powerful technology: deceleration through prohibition, domination that concentrates power in one organization, and doom. His diagnosis was that deceleration is already impossible and domination is becoming the world's default path. He warned that security where only I hold the tools eventually invites backlash from others and an arms race.
He presented defensive decentralized acceleration as the alternative and emphasized a security philosophy that weighs prevention over after-the-fact deterrence. To the long-standing security question of whom to trust, he named open source, formal verification, minimal trusted dependencies, and cryptographic proofs that reveal only the necessary facts as answers, and he saw zero-knowledge proofs and blockchains playing a role here.
As the first areas that must be trustworthy, he named operating system projects such as GrapheneOS, Qubes OS and formal verification of kernels. He said a TV or hardware wallet whose behavior cannot be known is unacceptable and called for more open and verifiable semiconductors. He also introduced a community-built open-source air quality monitor and a pilot project that studies air data while protecting privacy with FHE.
He said today's social media is polluted by bots and vested interests and is nearly useless for contentious issues or investment decisions, and he called for decentralized social media with interoperability and user control. Messengers need open source and end-to-end encryption plus metadata protection, and for AI he pointed to reducing the privileges of external models, such as filtering personal information with a local model before using a remote frontier model.
Finally he urged treating data as hazardous material by default and not collecting it at all where possible. He said neither trusting no one nor trusting big corporations or charismatic influencers who stand against them is the answer, and he ended the talk by saying the answer has to be found together.
Vitalik Buterin · Audrey Tang · 11.03
A conversation with Audrey Tang, Taiwan's first Minister of Digital Affairs. It began with the point that Ethereum's first task is to survive, and that its core is security, privacy and open source.
When Audrey Tang asked about priorities for the next ten years, Vitalik Buterin named survival. He answered that institutions value Ethereum not for speed but for a chain that has never stopped and its layers of defense, and that protecting the core properties of security, privacy and open source always comes first.
As something to stop doing, he named the attitude of clinging to making the world 5% better. He said an app that reduces the probability of losing all your principal matters more than a finance app that raises a 4.2% yield to 4.8%, and that priorities should be reset toward building simple things properly.
Admitting that participation in core research and development is still largely a social game, he saw that problems with measurable metrics, like a monthly AI competition, open a path for anyone to contribute to core development. Privacy protocols are scattered because goals differ, such as confidentiality, anonymity and metadata protection, and he hoped for a solution that lets people choose trade-offs like a slider instead of fighting between camps.
Asked how to verify that an agent running a treasury kept not only the written rules but also their intent, Vitalik saw this as one case of the alignment problem and said there is no easy answer. The key is who sets the sandbox rules around inputs, outputs and model replacement, and he expected good practices to accumulate after many experiments.
He said the boundary has been drawn not by philosophers but by compromises among forces worn out from long disputes, citing the Magna Carta and the Peace of Westphalia. He answered that what we can do is actually build technologies and institutions that provide both good transparency and good privacy, so that they are reflected in that line.
Many audience questions were light ones, such as why he wore a mask and which phone he uses, and Vitalik answered that he uses a Pixel 10 running GrapheneOS. He advised using vibe coding to broaden understanding rather than to replace it, and talked about the rapid change in Mumbai, which he visited after 10 years, and the higher level of crypto discussion in India.
As what he is most excited about, he named AI bringing forward cryptography once thought to be decades away, such as the zero-knowledge proof singularity and obfuscation. As a worry, he cited the burning house meme, saying that the fire is producing good things while burning the house down, so the house must be protected first.
Vitalik Buterin · Nicolas Consigny · 11.03
He introduced the Kohaku SDK and CLI, which help wallets provide privacy without extra setup, and ran a live demo.
Nicolas Consigny introduced Kohaku as an SDK that bundles strong privacy and security features, a reference implementation for power users that shows those features, and collaboration with other wallets. Currently ready are Railgun, Tornado Cash and Privacy Pools packages, a provider containing a light client, and a social recovery kit built by Wonderland.
He explained that in the beta extension demo, putting funds into a private pool was no different from a normal transfer, and that this was exactly the goal. He also demonstrated a Tornado Cash deposit made in the worst conditions, on hotel Wi-Fi and a VPN, and a CLI that processes several transactions in a batch.
Private reads, which keep node operators from seeing what is queried, are being developed along two tracks: an ORAM approach inside a TEE and a PIR approach. The former is limited because nodes with a TEE are rare, and the latter needs expensive GPUs. Vitalik Buterin saw out-of-protocol reward schemes as easy to abuse, and thought a structure with a free tier, above which users pay directly like a transaction fee, is realistic.
Next-generation privacy protocols, better social recovery, and network-layer privacy better than the already integrated Tor were named as the next challenges. Vitalik said to first firm up security and polish, and to make the Ethereum Foundation using it for its own internal operations the first milestone, and he reported that integrations with several wallets are also in progress.
Asked whether the one-click migration button should be postponed until frame transactions arrive, Vitalik answered not to wait. His position is to release the quantum-resistant migration that is possible now, relying on bundlers for the time being even if decentralization suffers, and to raise the guarantees as the technology becomes available.
Vitalik's diagnosis is that in a world where anyone modifies code with AI, a thoroughly verified security core becomes more important. He said he has become more optimistic over time because a CLI is easy for a bot to operate or to put a UI on top of, and a proposal to fine-tune a small model specialized for Ethereum and security work also came up.
He answered that the real volume of private transactions opens up when hardware wallets properly support privacy protocols, and that for now only Railgun is supported. As the biggest misconception he named the expectation that Kohaku will release a wallet itself, and he added that AI threatens privacy but can also be used for risk warnings and stronger anonymity.
Justin Drake · Kev · 11.03
Saying the point at which quantum computers can break cryptography has moved earlier and that AI may find new ways to break it, he presented Mission 2028, which proposes moving all of Ethereum to quantum-resistant methods by December 2028. The Ethereum Foundation's official target is December 2029, so this is closer to a proposal to bring it forward than a change to the official schedule.
Justin Drake said vulnerabilities found with AI have grown explosively, and that since March he has passed 895 potential issues to client teams, of which 500 were confirmed as real vulnerabilities. He drew a line, however, saying the topic of this talk is not individual bugs but the cryptographic assumptions the whole L1 relies on.
Drake sees Q-day in 2032 but also puts a 10% probability on it arriving by 2029, so he said we must hurry. Large investment is flowing into neutral-atom companies, and every layer of quantum technology is speeding up, for instance an amateur group cut the qubits needed for Shor's algorithm by 30% and shortened run time to a third. He also previewed a paper soon to come that lowers run time from minutes to seconds.
He explained that as quantum research became a national security matter, research was constrained, for example the Google paper was published a year late with its circuit hidden behind a zero-knowledge proof. The more underestimated threat is the new mathematics AI produces, and he warned that when conjectures held true for decades are overturned by counterexamples AI finds, Ethereum's cryptographic assumptions cannot be an exception.
Accordingly, he announced that the team will greatly pull forward the Strawmap and finish quantum-resistant transactions on the execution layer, quantum-resistant attestation voting on the consensus layer and quantum-resistant sampling on the data layer all by the fork after Hegotá (not yet named, so called J*). The schedule is to finish the transition by December 2028 with three forks over the 24 months after Glamsterdam mainnet in January 2027, and leanVM, which supports it, has reached the milestone of end-to-end formal verification.
Kev then explained that all components will be stabilized by the end of 2027 and then added to later forks in turn. They chose hash-based cryptography as the most conservative option, and moved the default hash from Poseidon2, used for performance, to SHA3/Keccak, which has had long analysis. The problem of growing signatures and proofs is solved by leanVM compressing many of them into one small proof.
He introduced that proving equivalence between the specification and the implementation keeps safety intact even as AI keeps optimizing the implementation, and that results such as a 4x improvement in proving speed and a halving of RISC-V cycles have actually appeared. But he cautioned that formal verification used for code deployment is still early and its trust base is not yet well understood, so automatically optimized implementations should not be shipped to production as they are.
Validators should change their 0x00 withdrawal credentials and take part in quantum-resistant validator registration, developers should find code that assumes ECDSA, and wallets should prepare to support frame transactions. Since 72% of the total ETH supply sits in accounts that have already exposed their public keys, he said that users who are worried can also move to a new account that has never signed.
Presentation video · Q&A video
Martin Hansen · Ethereum Foundation · 11.03
He estimated what counts as core infrastructure essential for Ethereum to run, such as clients, and how much neutral funding it needs at minimum.
Martin Hansen of the Ethereum Foundation's Funding Coordination team said this work started from a conversation on a green bench in Buenos Aires (presumed to be Devconnect) a year ago. The kernel refers to the work everyone relies on so that Ethereum stays alive, works correctly and can be used without permission.
Neutral funding with almost no strings attached prevents sudden project shutdowns and makes long-term work possible without chasing quarterly results. The more critical the infrastructure, the fewer sources there are that fund it without compromise, so the speaker stressed that the burden the Ethereum Foundation and public goods funders have carried so far should be shared by the many organizations that rely on Ethereum.
There are two criteria. Does stopping the work threaten liveness, correctness, or permissionless access? And does it lack a commercial motive, so that it risks being neglected or captured by a particular party? Quantum resistance research meets both and enters the kernel. zkEVM is left out for now because the network would not halt if it stopped today. Staking operators are excluded because they already have a revenue model in rewards and fees.
The map was drawn through 40 to 50 conversations with more than 25 experts. The protocol layer includes clients, protocol specifications, testing, protocol security, bug bounties, resilience research, and core dependencies. The access layer, whose definition is still less established, includes the language stack, trustless access, and risk transparency.
Using Protocol Guild compensation data, the team calculated the required staffing from scratch and estimated that keeping the kernel running at a minimum costs $14.7 million per year. This is a floor, not a budget, and actual spending will be higher. It will be reviewed every 6 months as Ethereum changes.
Because splitting incoming money evenly across all functions could leave every one of them below its minimum, the speaker proposed a waterfall approach that fills needs in order of priority. The order is Run, which keeps the chain operating to specification, Last, which carries it through long downturns, and Reach, which widens permissionless access and risk disclosure. Eligibility criteria for each project and gathering community input remain open tasks, and a related post is planned for the Foundation blog.
Yann Vonlanthen · 11.03
Yann Vonlanthen gave a brief overview of a staged roadmap to cut Ethereum's finality time, currently about 16 minutes, to a few seconds. Beyond consensus decoupling and one-round finality, he presented new research on a safety mechanism that speeds up finality while raising economic security to as much as 100%.
Finalizing an Ethereum transaction currently takes about 16 minutes, but the Strawmap targets finality in seconds. This requires effectively redesigning proof-of-stake Ethereum, and he explained that the changes will come in stages rather than all at once.
The current consensus engine, Gasper, intertwines LMD GHOST and Casper FFG. If the two are separated so that the available chain and the finality gadget each run at full speed, fast confirmation comes first and accountable safety, meaning economic security, follows after about 4 minutes. He stressed that EIP-8453, which specifies this, is a proposal to replace consensus wholesale.
Skipping the justification step and finalizing directly cuts the time to finality by about 50%. The trade-off is that signals arrive faster but accountable safety drops somewhat.
To make up for this loss, he introduced new research on a third component that adds no votes or network load and only changes how blocks are interpreted. With it, finality gets faster while economic security rises from today's 33% to as much as 100%, and reversing a safe block would require far more stake to be slashed.
Going below the minute scale requires networking improvements, validator consolidation, and shorter slots, and how these pieces fit together can be tried out directly on finality.watch. In the Q&A he mentioned networking proposals such as EIP-8334 and a redesign of the gossip layer, and said the theoretical limit of the current structure is two slots: confirm in one slot and finalize in the next. He repeatedly urged validators to consolidate.
0xRajeev (moderator) · Vitalik Buterin · Fredrik Svantes · Mooly Sagiv · Josselin Feist · 11.03
This panel covered how to keep Ethereum secure as AI advances faster than existing systems can be fixed. Fredrik Svantes of the Ethereum Foundation described the response after bug reports rose from 5 a week to 150 a day. Mooly Sagiv of Certora (presumed) and Josselin Feist of Aave (presumed) discussed the uses and limits of formal verification, and Vitalik Buterin discussed full-stack security and agent sandboxes.
Vitalik said he is optimistic about the end state of security, since math problems have lower bounds and a program has a finite number of bugs. However, he expected that AI and offensive capabilities will advance faster than existing systems can be fixed, making the next few years a hard period in which Ethereum and the whole internet must hold out with everything they have. Fredrik added that the habit of hiding security fixes in ordinary commits no longer works against LLMs.
Sagiv recalled that in academia he moved between papers and lost out to an Australian team that stayed focused on a single project for a long time. He argued that the Ethereum ecosystem is likewise losing depth by trying to cover every language, tool, and property. His position is that it should pick where to fight, such as the safest consensus protocol or wallet, and dig in all the way.
Fredrik said that around January this year, external vulnerability reports rose from about 5 a week to 150 a day, and 99.95% of them were useless AI-generated output. The team brought this back to a manageable level by requiring reporters to burn $10 worth of ETH, standardizing the reproduction environment and PoC format, and building a triage pipeline in which agents verify reports and cross-check them against other clients. The remaining problem, he said, is that agents tend to inflate severity to please the reporter.
Sagiv said modular design and early adoption are the key, and that the bear market, when audit costs are hard to afford, opens an opportunity to do formal verification with LLMs and tools. Feist described spec-driven development as still experimental, and drew a line by saying that correct reasoning on top of a wrong specification leads to wrong conclusions, so formal verification is a means of raising confidence rather than proving that bugs are absent. Vitalik stressed that the full stack must be considered together, since recent large losses came from UIs and centralized components rather than smart contracts.
Vitalik proposed a sandbox that layers hard rules, such as a $1,000 daily limit or human confirmation for anything over $100, with soft rules in which another AI judges the risk. He said there is no single solution because the sources of risk vary, from malicious apps and prompt injection to backdoors in the model itself. Fredrik predicted that before long people will use Ethereum through agents instead of browser extension wallets, and that the meaning of work such as verifiable frontends may change as a result.
Vitalik put the probability that banks will come to require in-person confirmation for large online transactions at 10%, and the probability of an incident within the next 10 years larger than all past hacks combined at 20 to 40%. He said an event in which the secrets of a huge number of people leak at once is more than 50% likely, and that the probability the Ethereum blockchain does not halt through 2030 is between 75% and 90%, which is hard to pin down. The other panelists advised using the slack AI creates on security rather than on release speed, and preparing for the worst case in advance.
Feist confessed to missing a bug after leaving understanding to AI, and recommended protecting expertise by explaining one's own understanding and taking the AI's rebuttals, rather than asking how a component works. Fredrik said open-weight models have increased the Foundation security team's ability to find vulnerabilities by at least 50 times, and Vitalik painted an uneven future in which a secure core that keeps hardening coexists with outer areas that keep getting hacked.
Austin Griffith · 11.03
BuidlGuidl founder Austin Griffith used experiments with his own bot to show how AI agents have changed the path from an idea to an Ethereum app. Introducing changes in education and tooling such as SpeedRun Ethereum and ETH Skills, he emphasized the parts that still require human judgment, such as incentive design and acquiring real users.
Griffith asked the audience whether they still read or write code by hand, and said hand coding has become closer to a hobby like throwing pottery in the basement. For him, code was a means of putting something into people's hands, and he said that now, if you have an idea, you can ship it right away.
He said that earlier this year he gave his bot a wallet, GitHub and Twitter accounts, and BuidlGuidl tools, and hundreds of thousands of dollars accumulated in the bot's wallet from fees on a token someone issued for the bot. The bot built a vesting contract and a burn mechanism, and also released an app in which an LLM trained on the bot votes on behalf of each person who stakes the token, raising DAO participation to 100%.
He said the service, in which several containers audit a contract for 1 USDC, handled 700 to 800 requests, and was meant to show that the era when an initial audit cost tens of thousands of dollars is over. Wedgie, a hardware wallet built from ordinary electronics-store parts and a 3D-printed case, and a wallet that plugs into a phone were also all built by the bot.
He said a single pass gets you to 80%, but the remaining 20% gets harder and harder. Citing a case of AI overimmersion in which someone built a giant dashboard that nobody used, he pointed out that what matters in the end is who actually uses it, how incentives are designed, and taste and distribution. He added the assessment that this is the best time to found a company alone but the hardest time for junior developers.
SpeedRun Ethereum, which used to take weeks, has become a Socratic tutor: you download the repository, hand the assignment to an agent, and it teaches by asking questions. ETH Skills, a knowledge file for agents, uses ETH Evals to filter out what a base LLM already knows and reduce wasted context, and capture the flag contests have also become events where each person brings their own AI to compete.
On an era in which agents write code on their own and persuade people, he answered that an instruction not to leak the private key cannot stop them, so things that must never happen should be left to the blockchain's enforced rules. He expected that agents writing Solidity contracts is becoming feasible with formal verification and audits alongside, and that someday it will be riskier for humans to write them directly.
Barnabé Monnot · 11.04
He explained, with a timeline, the order in which block time and finality will be reduced.
Barnabé Monnot introduced the goal of the organization he co-founded (presumed to be Ethlabs) as making Ethereum the settlement layer of the world economy, with ETH at its center. He explained that bearer assets, where holding is owning, and also assets registered elsewhere gain composability and availability when connected to an open chain that does not stop.
He defined utility as the value of a service minus its cost, and pointed out that if Ethereum is slow, users will choose less secure alternatives. Noting that many bridges have been hacked because they did not wait for Ethereum finality and relied on a group of reporters that is effectively a multisig, he stressed that getting faster means lowering the cost of the safe choice.
The goal is to cut the 12-second slot to 10 seconds in 2027 and keep lowering it gradually to 2 seconds. He explained that more frequent blocks improve user experience, update onchain prices faster so less value leaks through arbitrage, and increase opportunities to propose blocks, which also raises censorship resistance.
The goal is to cut finality, currently about 15 minutes, to 1 to 2 minutes by 2028 and then to 10 to 20 seconds. In the meantime, the fast confirmation rule, which is already running, provides confirmation within 13 to 24 seconds, slightly weaker than finality but 30 times faster. It has been applied to the Gnosis chain and is slated to be included in the Chainlink bridge package. He added that even when assets leave L1, their roots stay on L1, so L1 becomes a better place to issue and hold assets.
Mindful of an atmosphere in which quantum and AI threats have become the topic of the day, he made clear that security is still the foundation of Ethereum. But because other chains will also prepare and keep shipping, he said responding to threats is only the baseline, and that a social attitude of moving with urgency but without panic is needed.
Thanks to AI, formal verification of the new consensus design was completed at a speed that would have been hard to imagine six months ago. He proposed shipping other goals alongside a quantum- and AI-safe L1. By 2028 he listed 8-second slots down from 12, finality in 1 to 2 minutes, 5x scaling over today, and privacy.
He answered that rollups are already at around 100 milliseconds, so the L1 cannot catch up, and that the L1 has other strengths such as distribution reach. On the concern that quantum-resistant signatures are 10 times larger, he said it would be addressed through network improvements and a consensus separation design. On the concern about block building centralization, he said it would be addressed by reducing the per-block gas limit while maintaining throughput. He considered L1 preconfirmations hard to implement on a permissionless validator set.
Cheeky Gorilla (Protocol Guild) · 11.04
A presentation that summarizes in one place what goes into the next two hard forks, Glamsterdam and Hegotá.
Chikki Gorilla of Protocol Guild compared the roadmap with the one presented six months ago and said 108 places changed in the meantime. About one third are wording fixes, but changes are indeed frequent. He argued that this flexibility is a strength, not a flaw, and is the foundation of the 100% uptime the network has maintained.
The focus of the next six months is Glamsterdam, the scaling upgrade. It centers on ePBS, which brings the separation of proposers and builders, previously handled outside the chain, into the protocol, and on block-level access lists, which declare in advance what a block will read and write so that parallel processing becomes possible. It also includes pricing adjustments such as raising the cost of permanent data, a cut of up to 71% in the cost of sending ETH to existing addresses, larger contracts, and faster withdrawals. The gas limit will rise from 60 million to 200 million.
A recent audit revealed that a single pricing adjustment could cause ETH to become locked in some contracts or, in the worst case, disappear, and the upgrade has effectively stalled. No funds are at risk right now, but the team is working out how many contracts are affected, and the release has been pushed to January next year. The explanation was that Ethereum does not ship first and fix later.
Hegotá, targeted for August 2027, focuses on censorship resistance and privacy. The core pieces are FOCIL, which forces a list of transactions that the next block must include, and frame transactions, which complete native account abstraction. Thanks to frames, people can live on USDC alone and privacy protocols can work without relayers. He also expected that validator quantum-resistant key registration, added to the roadmap on November 2, could become a third core feature.
He stressed that forks after Hegotá are still at the research stage and will inevitably change, but the end goals of a fast L1, a gigagas L1, a teragas L2, a quantum-resistant L1, and a private L1 have stayed the same for years. Once mainnet at 10,000 transactions per second and L2s at 10 million per second are reached through snarkification, which uses ZK proofs to make verification lightweight, the protocol enters maintenance mode. The only difference from the slide of a year and a half ago is that privacy is now stated explicitly.
He answered that the Protocol Guild roster currently lists 172 core contributors spread across about 16 independent organizations. He said the multi-client principle might become unnecessary if formal verification advances as hoped, and described rollups as taking on uses that do not fit mainnet, making Ethereum like a Swiss Army knife.
Tomasz Stanczak · 11.04
Stanczak, who stepped down as Ethereum Foundation co-executive director and now leads a robotics company, described his idea of using Ethereum as a coordination layer for AI operating in the physical world.
Arnav Bansal asked Tomasz Stanczak, who talked about the intersection of Ethereum and robotics at Devcon two years ago, about the shift since then. Stanczak said he had been drawn since around 2017 to a vision such as robotaxis that pay for and manage themselves, but that this had long been impossible and has become possible only now thanks to AI.
He said that while leading the Ethereum Foundation, he judged it unhealthy to push the Foundation too far toward AI and robotics. Complete Robot, which he founded after leaving the Foundation, is a subsidiary of Nethermind. He said he wrote most of the early Nethermind client code himself and is now repeating that way of learning from scratch.
He painted a future in which the world increasingly works like one giant machine and even cities are laid out like prefabricated structures, and set out a coordination layer for physical AI worldwide as his vision. For now the company is focused on building profitable data centers and on robot distribution, and he said a robotics team joined in October and opened a lab.
To a younger generation that seems to have lost its bearings in the age of AI, he recommended starting with problems close to home, such as family and the local community. He thought that in India, which has many engineers, people could learn quickly with AI and contribute right away to improving efficiency in local healthcare, logistics, and agriculture.
He said the most enjoyable thing he has learned recently was going back to physics fundamentals such as power engineering and electromagnetism, and urged engineers to study HVAC, civil engineering, and electromagnetism. In his words, it is not only useful but fun enough to get absorbed in.
Vitalik Buterin · Siddharth Shetty · Denver (host) · 11.04
Vitalik Buterin and Siddharth Shetty of Networks for Humanity, who has taken part in building India's digital public infrastructure, talked about building systems used by hundreds of millions of people. The main topics were a reflection that people problems were harder than technical ones, the design principles shown by DigiLocker and UPI, Vitalik's interests after the protocol cleanup, and the shift from organizational logic to coordination logic.
Vitalik said that things he expected to be possible only 25 years from now, such as the proof singularity, advanced cryptography, and privacy, are now 1 to 2 years away. He also said that security councils, which L2s have failed to shed for over five years, now look like an easy problem thanks to advances in formal verification, and that in an era when AI breaks everything, the value of Ethereum, designed not to break, has grown. Shetty said that as AI takes over cognitive labor, he looks forward to seeing what people will create from their inner lives, such as curiosity and emotion.
Vitalik answered that the hardest problems were always between people, not technical. On the technical side, the answers are now obvious, such as a 64-bit VM instead of 256-bit, an encoding better than RLP, simple proof of stake from the start, and skipping uncle blocks and the MODEXP precompile. But he said it is still hard to say what the right answer is for relationship problems in the ecosystem. He also said that 12 years ago, learning existing protocols and cryptography papers was itself hard, and that this frustration was a major reason he wrote his blog.
Shetty introduced India's digital public infrastructure, which started from the problem of bringing 1.3 billion people into the formal economy, and used DigiLocker as an example, which holds more than 9 billion electronically signed certificates. In a country of extreme diversity, a single app cannot solve everything for everyone, so with minimal intervention the ecosystem on top was left to solve problems, and UPI grew digital payment merchants from 3 million to 60 million. He also pointed out that policy moved alongside, for instance by changing the law to give digital certificates legal recognition.
Vitalik advised not to fix the solution first but to find the security, privacy, and interoperability properties that are lacking now, and then pick means such as zero-knowledge proofs or recording data hashes. Shetty said they are building a distributed directory that verifies state changes in public information, such as license revocation, and records hashes on Ethereum. He expected demand to arise in cases like an agent spinning up a smart contract escrow when it meets a seller it cannot trust.
Vitalik summarized the Lean roadmap in three strands: quantum and AI readiness, security including privacy, and scaling and efficiency, plus cleaning up cruft. He said that once these are done, attention shifts to the application layer. He listed questions such as metadata and network-layer privacy, oracle security, and whether the concept of an application is still needed in a world where anyone can place orders through smart contracts and agents read the conditions.
Shetty introduced Namma Yatri, a P2P mobility network that returned 250 million dollars directly to drivers, and an energy network that lets farmers sell surplus solar power to manufacturers in other regions. He said the next goal is to extend coordination logic, based on incentive design and cryptography, to areas such as drug discovery and nuclear reactor construction, in place of the organizational logic that solves problems by building institutions and hierarchies.
On the risk of a few people commanding huge swarms of agents, Vitalik answered that it is a problem for all of humanity, not just Ethereum, and that Ethereum should solve the parts it can. He also mentioned examples of d/acc in biotech that India took part in, such as the open-source vaccine PopVax. Shetty said India, having no legacy, has a big opportunity to leap ahead, and urged the younger generation to practice pragmatic idealism and to take a long view.
Toni Wahrstätter · 11.04
Toni Wahrstätter of the Ethereum Foundation summarized the last two years of L1 scaling and the plans ahead. Under the principle that scaling must not push full nodes and validators into data centers, he covered the three mechanisms that raise the gas limit to 200 million in Glamsterdam, ideas for after that, and problems still unsolved.
When the gas limit rises from 60 million to 200 million with Glamsterdam, 68 million simple transfers per day may seem possible. But under the EIP-1559 design, average usage is half the limit, and real transactions use more gas. Assuming 100,000 gas per transaction and 5 transactions per person per day, the chain fills up if only 1 in 9 Mumbai residents uses it, which showed why scaling is needed.
He explained that scaling is more accurately measured by useful work done per second than by TPS or the gas limit. Raising hardware specs makes it easy to increase capacity, but he made clear the principle that it must not go so far that full nodes and validators need data centers or GPUs, and that they must be runnable at home.
Blobs in EIP-4844 two years ago, then PeerDAS and blob-parameter-only forks raised the blob count to 21, and scaling on the L2 side advanced greatly. L1 execution was weaker, but a campaign to raise the gas limit prompted client teams to focus on performance optimization, and he recalled that the limit was doubled from 30 million to 60 million without adding protocol features.
ePBS (EIP-7732) separates the consensus block from the execution payload, extending the time available for execution from 2 to 3 seconds to about 6 seconds, and block-level access lists enable parallel execution across multiple cores. The repricing that sharply raises the cost of creating state is not a cure-all, but he explained that it buys time until the state problem is solved at its root.
Options being discussed include sending the block-level access list before the payload to gain one more second, or speeding up payload propagation. On the state side, the team is exploring a UTXO-style payment lane specialized for payments that barely grows state, cheap state stored on HDD instead of SSD, and keeping frequently used state in cache between blocks to add 20 to 30% efficiency.
A zkEVM, in which a single prover proves block execution and the other nodes only verify it cheaply, is the most powerful scaling tool, and a split binary tree (EIP-8297) that cleanly partitions state is also in preparation. EIP-8288, which detaches signatures from transactions and bundles them into a STARK, is a proposal like a beefed-up SegWit, and he introduced it as helping both scaling and the quantum-resistant transition.
He named a sustainable state solution, still-high proving costs, and the complexity of repricing that ripples out to applications as unsolved challenges. In the Q&A, he said that L2s are needed to serve the whole world but the L1 must also scale for users who trust only the L1, and he answered that hot state and cold state have different prices, so they should be set by the protocol specification rather than at client discretion.
Leandro Ferrigno · Ivan Litteri · 11.04
Leandro Ferrigno and Ivan Litteri of LambdaClass summarized the present state of Lean Ethereum as seen while building the execution client ethrex and the consensus client ethlambda. Lean Ethereum is not a single hard fork but a process in which gradual mainnet changes meet a new stack being refined on a quantum-resistant devnet, and under the revised roadmap, lean features enter starting with the fork after Hegotá.
The talk opened with the background that the Beam Chain, proposed by Justin Drake at Devcon Bangkok two years ago, was renamed the Lean Chain, and that in the meantime the threats from AI and quantum computing have grown much larger. Today's Beacon Chain uses BLS signatures for validators and ECDSA for users, and with Gasper, which couples fork choice and finality, it reaches finality in about 16 minutes.
Lean aims for finality within one minute, and eventually within seconds, using quantum-resistant hash-based signatures and a new consensus algorithm. Extremely Lean, still at the research stage, is a design in which each validator stores its own state and only proofs are posted to the chain. The presenters stressed that swapping the signatures alone would give quantum resistance, but this work is also a chance to pay off long-accumulated technical debt.
Hash-based XMSS, which is easy to aggregate, is being considered for validators, and a hash-based signature candidate for execution layer users. A hash-based approach is also being prepared for data availability, and leanVM, a minimal zkVM, gathers all of these into a single proof. Consensus is moving toward separating fork choice votes from finality votes, and most of the work is going through formal verification.
One is a greenfield quantum-resistant devnet in which several client teams took part and tested the Lean features all at once; the per-feature devnets have been run up to the fifth. The other is the incremental path of folding in one feature per mainnet fork. The greenfield path is fast and free of migration worries but hard for existing clients to follow, while the incremental path is slow but safe because the procedures are familiar.
The original plan to include Lean features by 2029 has been pulled forward significantly to fit Mission 2028. Hegotá is the last regular fork without Lean features and introduces frame transactions to prepare for swapping out cryptographic schemes. The next fork (placeholder name I*) brings leanVM and hash-based user signatures, and the one after (J*) brings validator XMSS signatures, data sampling, and a new polynomial commitment.
The team has released ethlambda, which started out on the devnet, for the Beacon Chain as well, and plans to test both experimental networks and mainnet upgrades by swapping modules within a single client. In the Q&A they said the structural change and hardware burden for L2s and home stakers should not be large, but that how to move from the Beacon Chain to Lean consensus has not yet been decided.
Aaryamann Challani · Rashmi Abbigeri · Simone D'Amico · Ruchir Dalmia · 11.04
Aaryamann Challani of EthSystems, Ruchir Dalmia, Polygon Labs head of Europe, and Simone D'Amico of ether.fi discussed, with Rashmi Abbigeri of DAOstar moderating, what keeps institutions from trading onchain. The discussion covered regulation, liquidity, fiat rails tied to bank business hours, the capital efficiency of atomic settlement, and differing views on CBDCs.
Ruchir Dalmia noted that the asset-side problems have largely been solved, but that regulation differing across markets and fragmented liquidity between stablecoins, tokenized assets, and fiat rails remain. He added that Indian institutions also cite regulation as a bigger obstacle than internal risk management.
Simone D'Amico said it is now time to prove capital efficiency with tokenized assets, and that they are building an experience in which companies earn interest on deposited assets, take collateralized loans without paperwork or permission, and even pay suppliers and payroll. He explained that making this so smooth that institutions do not even know they are using a blockchain is the job of app developers.
Aaryamann Challani explained that lower costs and an uptime record with no halt since 2015 are strengths of public Ethereum, but that a structure in which every transaction is public is an obstacle because of the custodial duty to protect customer information. He said modular privacy is needed, where the disclosure scope is chosen per flow; for a bond, for example, the ID, issuance amount, and maturity date would be public while the trade history and resale records would stay hidden.
Dalmia said that although onchain settlement is instant, institutional balances held in fiat remain tied to bank business hours, so the idea that institutional trading runs around the clock is a misconception. D'Amico lamented that even non-custodial setups must follow the same regulations and that crypto card issuance and UPI integration are blocked in India, while Challani named the lack of standards and specifications as the problem institutions raise most often.
It was said that institutions are reluctant to work together on shared infrastructure and that their calculations differ: traditional banks aim at deposit tokens, while fintechs aim to expand market share. Compatibility among token standards with differing rules and execution methods was raised, along with the role of liquidity providers who connect various forms of money. D'Amico described the effect of atomic settlement with a case in which Aave V4 was deployed directly and all supported assets could be used as collateral without permission.
The panelists expected real integrations and full-scale services such as OTC desks to increase beyond pilot projects, and D'Amico previewed payment links that carry value in a single link. On CBDCs, D'Amico argued they should be issued on Ethereum as a cash-like tool that cannot be frozen, while Dalmia countered that the case is weak in India, where digital payments are already well established.
Dalmia hoped that non-dollar stablecoins would gain strength in emerging markets, and D'Amico urged everyone not to forget why they started with blockchain even while adapting to institutions and regulators. Challani summed up that institutions do not want total anonymity but place high value on the cypherpunk values of censorship resistance, openness, confidentiality, and security.
Shankara Pailoor · 11.05
Shankara Pailoor, chief research officer at Veridise, outlined which attacks Ethereum's applications, rollups, data availability, and consensus layer would each be exposed to once a quantum computer that breaks elliptic curve cryptography exists. He presented scenarios such as quietly draining dormant accounts or collecting validator keys over a long period and spending them all at once, and stressed that in the end the time it takes to migrate is what matters.
After explaining qubits, error correction, and the difference between logical and physical qubits, he pointed out that what quantum computers do especially well is find the period of a repeating pattern. Shor's algorithm solves integer factorization and discrete logarithms, breaking RSA, ECDSA, BLS, and KZG, while hash-based and lattice-based schemes remain safe for now; the problem is that Ethereum uses elliptic curve cryptography in every layer.
He explained that as algorithmic improvements reduce the number of qubits required and hardware qubit counts grow, the two lines are converging. If migration time plus the required secrecy period is longer than the time remaining until Q-day, it is already too late; since onchain data is public by nature, he stressed that on Ethereum the time migration takes is the key factor and that the entire stack must be replaced.
The moment an account makes its first transaction, its public key stays on chain forever, so an attacker can harvest public keys and compute the private keys. The first attack may look like quietly draining dormant accounts holding a lot of ETH, and the top 1,000 accounts with exposed public keys hold 20.5 million ETH. He also cited an estimate that if admin keys holding mint and burn authority are stolen, about $200 billion in stablecoins and real-world assets would be at risk.
For ZK rollups that use pairing-based SNARKs, if the setup secret is recovered, a fake state transition proof that takes the entire bridge funds could pass, so everything up to the final wrapping step must be handled with STARKs to be safe. The KZG used in PeerDAS is also built on a secret from a ceremony with more than 140,000 participants that a quantum computer could compute directly, which makes possible an attack that hides data while passing sampling checks and blocks fraud proofs for optimistic rollups.
If an attacker seizes validator BLS keys, controlling more than one third of stake can halt finality, more than half allows block reorganization and censorship, and more than two thirds allows finalizing any chain they want. It is the hardest target because many keys must be broken one by one, but he warned that validator keys rarely change, so they could be quietly collected for about a year and spent all at once, and that validator consolidation is also reducing the number of keys.
NIST has released lattice-based and hash-based standards, and Ethereum researchers are reviewing alternatives for every layer, but he said safely moving the whole ecosystem to the new cryptography is hard. In the Q&A he answered that Ethereum prefers hash-based schemes for their simple structure, and that privacy protocols using elliptic curves face a big task in putting quantum-resistant cryptography inside ZK circuits. Ordinary users can reduce their risk right away just by moving to a new account.