Decentralisation Accelerationism (d/acc)

d/acc is a technology philosophy coined by Vitalik Buterin in late 2023, proposed as an alternative to both AI doomerism (halt dangerous technologies) and e/acc (accelerate unconditionally). The “D” carries three simultaneous meanings: defensive (technologies that protect against threats), decentralized (no single actor controls critical infrastructure), and differential (selectively accelerate beneficial technologies, especially those that are symmetric). (→ [[devconnect-argentina]])

The core test: “Would you be happy if your worst enemies also had this technology?” Technologies that pass — vaccines, clean air monitoring, open silicon, privacy tools — are d/acc. Technologies that create asymmetric power advantages (surveillance, offensive drones, closed AI inference) are not, regardless of how decentralized their infrastructure is.

d/acc treats open source as a default condition rather than a preference. Biotech, silicon, and pharmaceuticals are identified as critical sectors where this norm is conspicuously absent and must be built from scratch.

Note: Albert Ni (Zerox Park) claims the concept was first articulated at Zerox Park’s “Prague Crypto” event before Vitalik’s blog post was published.

Cryptographic Privacy Stack

The deepest technical cluster at the Buenos Aires dacc track. Consensus: no single cryptographic primitive solves privacy; TEEs, FHE, ZK, and ORAM must compose.

Threshold FHE Breakthrough (Anonymous Researcher)

From 10 decryptions/second at 500ms latency → tens of thousands per second at <10ms. ~1,000× improvement on the same hardware versus TFHE standard. Key mechanism: new parallelization scheme for threshold decryption. Opens the path to:

  • Private AI (small language models computed under FHE without TEEs)
  • Private DeFi at Visa-scale throughput
  • Encrypted smart contract state that cannot be read by validators

This aligns with the Fhenix breakthrough reported at ECC2 (→ Privacy as UX Design).

ORAM — Oblivious RAM (Afonso Tinoco / Oblivious Labs / CMU)

TEEs encrypt content but leak access patterns: an attacker observing memory reads/writes learns which data is being accessed even without decrypting it. ORAM (Oblivious RAM) reshuffles data on every access, making all access patterns identical regardless of the actual query.

Signal precedent: switched from linear-scan private contact discovery to ORAM, reducing infrastructure from ~500 servers to 6 (deployed 2022). Most TEE deployments still do not use ORAM — they leak metadata even with encrypted content.

Latency comparison by method:

  • ORAM inside TEE: microseconds
  • MPC-based oblivious access: ~30ms
  • ZK PIR (private information retrieval): ~12ms

Work in progress: open ORAM standard library; oblivious Ethereum EVM (estimated 3-year timeline).

Adversarial Interoperability via Confidential VMs (Andrew Miller / Teleport)

Users have a legal right to their own data on platforms like Twitter/X. Practical problem: API access is restricted and revocable. The “OAuth3” paradigm: run your logged-in browser session inside a TEE (confidential VM). The TEE signs an attestation that it ran your session authentically. Extract your behavioral data without platform permission. Delegate to third-party apps that also run inside TEEs — apps that inherit your authenticated access without ever seeing your credentials.

This is adversarial interoperability through cryptographic coercion: platforms cannot block access without blocking all TEEs, which would conflict with cloud providers’ GDPR compliance obligations. GDPR compliance is accidentally funding TEE infrastructure useful for web3.

TEE vs. FHE vs. MPC: The Pragmatist Position

Debate (Andrew Miller, Justin Glibert, Afonso Tinoco — panel):

  • TEEs: deployable today; trust hardware manufacturers (Intel/AMD) and cloud operators. Acceptable because: cloud providers won’t burn zero-days on small crypto apps when it risks GDPR clients. Pragmatic near-term choice.
  • FHE: mathematically stronger; requires MPC for decryption (same trust model as MPC); scaling rapidly. Medium-term (2–5 years).
  • MPC: same trust model as FHE at decryption; currently slower than TEEs.
  • IO (Indistinguishability Obfuscation): would eliminate need for MPC at decryption entirely; theoretical attack succeeded but was an implementation flaw, not a fundamental break — research ongoing.

Key insight: “Local privacy” (Zcash, Tornado) = zero-trust but non-programmable. “Global privacy” (FHE-encrypted state) = programmable but trust-minimized. Programmability beats theoretical purity in practice because impractical systems get abandoned.

Open Hardware and Silicon

The access gap: designing one chip costs ~$300K/person/year in EDA tooling alone. NDAs are required to access chip IP, making it legally impossible to publish security fixes. The lowest available open process node is 130nm; the lowest closed is 2nm — a ~65² transistor-density gap.

FHE-specific constraint: FHE acceleration requires chips with gigabytes of SRAM (current commercial processors have tens of megabytes). No open-source chip exists that meets this spec.

Vensa nonprofit (M Pang, Vitalik-backed): Goal: democratize chip tape-outs via:

  • Free shuttle runs (shared MPW slots on open process nodes)
  • Open IP library
  • Open PDKs (process design kits)
  • Accessible tape-out pathways for students and researchers

Fabric Cryptography (Michael Gao): Building a custom FHE accelerator chip targeting inference-speed encryption of frontier LLMs. Gen 1 chip: releasing within one year of the talk (late 2026). The business thesis: privacy needs custom silicon because current TEEs are controlled by 3–4 companies and cannot be trusted for high-stakes applications.

Trustless TEE Initiative (Quintus Kilbourn): Current hardware security relies on manufacturer goodwill, not cryptographic guarantees. Three-year goal: produce an open, physically-secure TEE with three properties — security, openness, verifiability. Hardware root-of-trust that is independently auditable.

uCritAir (M Pang + Justin Glibert): Open-source gamified air quality monitor that runs a full ZK STARK (Merkle inclusion proof with nullifier for anonymous attestation) plus lightweight FHE encryption on a $3 microcontroller. Demonstrates that FHE does not require data-center hardware. Key design:

  • ZK STARK for device identity (rotating weekly nullifiers for unlinkability)
  • Lightweight FHE encryption on-device before transmission
  • Encrypted data lake with scheme-switching to more powerful FHE for computation
  • Differential privacy noise added pre-encryption

dGEN1 Phone (Markus Haas, Nicola Ceornea): Android AOSP phone with embedded Ethereum wallet, Helios light client, and account abstraction wallet on a MediaTek chipset. Hybrid architecture: closed-source SOC (pragmatic) + open-source TEE for high-value computation. Threat being addressed: Google’s planned 2027 requirement that all Android app installations be signed by KYC’d developers — effectively ending sideloading, described as “a nuclear bomb for privacy.”

Biological and Environmental Defense

The most surprising cluster at the track. Physical-world defensive technologies applying open-source and decentralized principles to domains controlled by incumbents with structural conflicts of interest.

Open Water — LIFU (Aaron Timm, Erin Magennis): Low-Intensity Focused Ultrasound — millimeter-precise, non-invasive brain stimulation via smartphone-registered MRI alignment. Results:

  • 10 sessions → 50% remission in treatment-resistant depression
  • 95% microclot dissolution in 2 minutes of sonication
  • Cost: $250K → $10K (after open-source design)
  • License: CC ShareAlike 4.0 + AGPL

FDA approval is frozen to a specific device version; open-source iteration resets the clock. This is structural regulatory capture, not accidental.

Long COVID Labs (Rohan Dixit): >100M long COVID sufferers; >$1T/year economic cost. Dixit self-cured with monoclonal antibody + Paxlovid combination. Raised $2.6M via community auction. 88,000 patient self-reports analyzed. Key structures:

  • IRB-approved blockchain patient registry with token rewards for self-reporting
  • World’s first patient-owned IP NFT issued by a DAO
  • Big pharma absent (unclear reimbursement pathways)
  • Vitalik identified as one of the few funders advancing this research

Science Integrity / Tamper-Proof Bioreactors (Erin Magennis): 70% of published scientific experiments have data manipulation or replication failures. The Alzheimer’s amyloid fraud (discovered ~2024, first paper 2006) wasted 18 years of research and billions in funding. AI scientists trained on fraudulent literature will compound fraud at scale. Proposed solution: open-source bioreactors with device-level signing — every experimental output is timestamped and cryptographically attributed to a specific device in a specific configuration.

Air Quality — CO₂ and Ventilation (Georgia Lagoudas / Air Club): CO₂ at 1,400 ppm → 50% reduction in cognitive function (measured). Building ventilation standards were set by smell (80% non-complaint threshold), not health outcomes — never updated after COVID. Interventions with measured effects:

  • Air purifiers → 8% test score increase (equivalent to halving class size)
  • Montenegro: national school air quality program
  • 1 COVID infection → average 3-point IQ drop (permanent)

Information Integrity and Coordination

Cognitive Sovereignty (Sam Gbafa / Tiny Cloud): The “exocortex” — external tools that extend cognition — is becoming part of self-identity. Platforms extract value by pushing users from neocortex (reasoning) to reptilian brain (reflex/fear). Evidence: math remediation enrollment at UCSD increased following Instagram Reels launch (2020). Cognitive sovereignty requires controlling digital identity via cryptographic keys — not through trust in platforms.

Agora Deliberation Tool (Yuting Jiang): Polarization creates centralized-power capture risk. Polis-style ML opinion clustering: 200 participants, 100 statements, 7,000 votes in a French transportation strike consultation. ZK passport (Semaphore-based) for anonymous verified participation. Funded via EU NGI grants. Shows institutional adoption path for ZK identity without doxxing.

Ghost Sexism (ml sudo): Deniable, indirect sexism — “ghost sexism” — is the primary remaining barrier for women in tech. Trustless systems remove the human gatekeeping layer where deniable discrimination operates. Historical analogue: married women’s property rights (Married Women’s Property Acts, 1839–1900) — the structural mechanism is identical. Women are 47% more likely to be seriously injured in car crashes because crash test dummy standards were set using male physiology only. AI trained on discriminatory data will amplify these outcomes.

True Markets (Tangle): Fully on-chain verifiable prediction markets as “civilization’s feedback layer.” Current prediction markets are semi-centralized (opaque oracles, permissioned liquidity). Architecture: Uniswap v4 hooks + yield-bearing collateral + AI oracle resolution. Futarchy integration for governance decisions. (→ Prediction Markets)

Geopolitical d/acc

Taiwan: Fabian d/acc (Ashm / Harvard Kennedy School / ex-Ministry of Digital Affairs): “Slow resistance, fast networks.” In Mandarin, “digital” and “defensive” are pronounced identically — not accidental. Taiwan’s digital infrastructure strategy: privacy-preserving digital identity wallets raise the cost of authoritarian one-shot information capture without provoking military escalation. Technology as societal hardening: make control expensive, not control impossible. Key insight: the goal is not an arms race but a deterrence that preserves ambiguity.

Vitalik’s Explicit Non-d/acc: Drones are “very not d/acc” despite being decentralized and accelerationist. AI agents similarly — currently more useful for offense than defense. The “D” genuinely constrains the frame: the question is always “does this create symmetric defensive value, or does it create asymmetric offensive advantage?”

Public Goods and Funding

Octant (Julian Zawistowski): Public goods funding is a governance problem, not a charity problem. Three unsolved sub-problems: allocation (incomplete preferences), deployment, and impact measurement. Mechanism design in web3 has largely failed at scale (ENS Harberger tax: designed, never shipped). Blockchain enables experiments, but experiments need evaluation infrastructure.

Bhutan Case Study (Chiag): National digital identity on Ethereum. Open-source soil sensors (NPK/moisture, LoRa WAN — no telecom required) for subsistence farmers. Open carbon credit platform on blockchain. Gross National Happiness as an alternative economic metric. Shows that d/acc principles map onto small-nation sovereignty concerns, not just privacy-focused individuals.

Connections

  • Cypherpunk Values & Philosophy — d/acc extends the cypherpunk tradition; shares the defensive privacy framing; distinct in explicitly including biotech and hardware
  • Privacy as UX Design — ORAM, threshold FHE, adversarial interoperability (OAuth3) directly complement the privacy UX stack
  • ZK Proving Infrastructure — FHE acceleration chips, ORAM for ZK provers, uCritAir ZK STARK on microcontroller
  • Ethereum Public Goods Funding — Long COVID patient DAO, Octant, tamper-proof science infrastructure; alternative public goods funding models
  • Prediction Markets — True Markets: Uniswap v4 hooks + AI oracle resolution + futarchy
  • Metadata Privacy — ORAM hides access patterns that TEEs leak; adversarial interop extracts data without metadata leakage to the platform

Open Questions

  • Can open silicon (Vensa, 130nm open PDKs) close enough of the transistor-density gap to produce viable FHE accelerators on open hardware?
  • Does the Trustless TEE Initiative produce hardware that institutional cloud providers will adopt, or only niche crypto deployments?
  • At what point does the “slow resistance” Fabian strategy (Taiwan framing) become insufficient against adversaries willing to move quickly?
  • Can patient-owned IP (Long COVID DAO) create a replicable template for decentralized clinical trials on conditions that pharma avoids?
  • Will the d/acc frame successfully recruit biotech and hardware communities that currently have no crypto connection, or will it remain a primarily crypto-native philosophy?