Research doc for the 2026 biohacking landscape overview. The author's #1 end-goal is substrate independence / "Friendship is Optimal" (FiO). This is the most important doc in the set, so it is written to be rigorously honest about demonstrated vs. speculative. Date of research: 2026-05-31.
Cross-links (do not re-derive here): - Brain-preservation chemistry & quality (aldehyde fixation mechanism, ASC protocol, scan-modality resolution ladder, ischemia, M22 vitrification):
/workspace/cryonics/cryonics-posts/and/workspace/cryonics/alcor-research/. - The connectome-insufficiency critique, Hayworth's challenges to Alcor, and the 2026 Fahy/Coles vitrified-human-brain study:/workspace/cryonics/alcor-research/research/alcor-news-bad.md,alcor-timeline.md,alcor-writing-catalog.md. - The AI dependency (FiO needs a superhuman AI to run/manage uploads):/workspace/cryonics/not-die/how-not-to-die-from-agi.md. - BCI as a possible bridge:03-bci-neuralink.md(this folder).
Claim: WBE was given its canonical technical framing by Anders Sandberg & Nick Bostrom, Whole Brain Emulation: A Roadmap, Technical Report #2008-3, Future of Humanity Institute, Oxford (2008), arising from a 2007 multidisciplinary FHI workshop. - Confidence: C1 - Sources: ORA/Oxford record; PDF mirror (gwern) - Date checked: 2026-05-31
The roadmap decomposes WBE into four requisite technologies: (1) scan the brain at sufficient resolution; (2) translate the scan into a computational model (segment neurons, infer parameters); (3) run that model on adequate hardware; (4) embody it (simulate a body + environment so the emulation has inputs/outputs). This decomposition is still how every serious actor frames the problem in 2025–26. - Source: NIMH summary of the four-technology framing echoes this; framework itself in the roadmap PDF. C1.
The roadmap's most durable contribution is that WBE is not one target but a ladder of "levels of detail," and we don't know which rung is sufficient to capture a person. The ladder runs (low → high resolution): computational module / brain-region connectivity → analog network population → spiking neural network → electrophysiology (membrane states) → metabolome → proteome → states of protein complexes → distribution of complexes → stochastic single-molecule behavior → quantum. The required level "is presently uncertain, and will determine the difficulty." - Confidence: C1 (the ladder & uncertainty are explicit in the report) - Sources: Roadmap PDF; level/FLOPS summary in AI Impacts wiki, citing Table 9 - Notes: This "which level is enough?" question is the unresolved scientific crux (see §5).
Claim: The roadmap's Table 9 estimates processing demand to emulate (run, not scan) a human brain at three levels the workshop deemed most plausible: spiking neural network ≈ 10¹⁸ FLOPS, electrophysiology ≈ 10²² FLOPS, metabolome ≈ 10²⁵ FLOPS. - Confidence: C1/C2 - Sources: AI Impacts wiki (Table 9); cross-ref researchgate "Prospects of WBE" - Reality check (C2, show work): 10¹⁸ FLOPS = 1 exaFLOP. Frontier (ORNL, 2022) and El Capitan (LLNL, 2024) are exascale machines. So the run-it compute for a spiking-level human brain has been available since ~2022. This is the single most important "what held up": the roadmap correctly predicted compute would not be the binding constraint — and it isn't. The binding constraint is getting the data out of a brain (scan + translate). The 2025 State-of-the-field report says this flatly: "The main barrier to better brain emulation models is more and higher-quality experimental data." (brainemulation.mxschons.com)
Connectomics = mapping the wiring diagram (every neuron, every synapse) via volume electron microscopy (EM) plus increasingly AI-driven segmentation. This is the genuinely fast-moving, demonstrated part.
Claim: The roundworm C. elegans hermaphrodite connectome (302 neurons, ~7,000 synapses) was completed by White et al. in 1986 after ~13 years of manual EM reconstruction — the first connectome ever. - Confidence: C1 - Sources: White et al. 1986 commentary, PMC; OpenWorm/Wikipedia - The lesson: OpenWorm (since ~2011) aims to simulate the worm cell-by-cell. After ~40 years with a complete connectome, there is still no agreed functional whole-organism emulation that reproduces real worm behavior from measured biology. The connectome lacks synaptic weights and signs (excitatory/inhibitory), which can't be read from a static map; behavior-matching simulations have had to tune/ML-fit weights rather than measure them. - Sources: Jefftk "WBE and nematodes"; LessWrong "No progress on C. elegans after 10 years"; functional-connectomics review PMC6630759. C2. - Memorable framing (C2): "Connectomes are the biological equivalent of what you'd get if you removed all the component symbols from a circuit schematic and left only the wires." (LessWrong)
Claim: The FlyWire consortium published the first synapse-resolution full adult fly brain connectome in October 2024 (Nature), from the FAFB (Full Adult Fly Brain) EM dataset: ~139,255 proofread neurons and ~50 million synapses, with 8,453 annotated cell types. - Confidence: C1 (two independent reports + primary) - Sources: Nature: whole-brain annotation & cell typing (Schlegel et al.); NIH Research Matters; Science/AAAS coverage - Lineage: Janelia hemibrain (partial, ~25k neurons, 2020) → FlyWire full female adult brain (2024). This is the first whole-brain connectome of an animal that actually does interesting behavior. C1.
Claim: The MICrONS program (Allen Institute / Baylor / Princeton) published (Nature, Apr 2025) a dense reconstruction of ~1 mm³ of mouse visual cortex: >200,000 cells, ~0.5 billion synapses, ~4 km of axons, co-registered with functional calcium imaging of ~75,000 neurons. - Confidence: C1 (two sources + primary) - Sources: Nature: functional connectomics, multiple visual areas; Allen Institute announcement; Princeton coverage - Why it matters: It pairs structure with function (which neurons fired to which stimuli) and is the largest mammalian volume to date. Field recognition: EM-based connectomics was Nature Methods "Method of the Year 2025." (Nature Methods) C1. - Scale anchor: 1 mm³ ≈ 200k cells. A whole mouse brain ≈ 70–100 M neurons → ~hundreds of mm³. A human brain ≈ 86 B neurons. MICrONS mapped roughly one-millionth of a human brain's neuron count.
Claim: The human brain ≈ 86 billion neurons and ≈ 100 trillion synapses (commonly quoted; synapse count is an estimate with wide error bars, often given as ~10¹⁴–10¹⁵). - Confidence: C1 for neurons, C2 for synapses (estimate) - Sources: UCLA BRI; Herculano-Houzel "86 billion and counting" PMC11884752 - Throughput / data reality (C2): Imaging a whole human brain "would require an estimated 17 million years by applying the same brain imaging protocol as was employed for Drosophila," and a complete human cortex at EM resolution "will require a zettabyte (1,000 exabytes) of data." (On human nanoscale synaptome, PMC11423976; big-data-of-connectomics PMC4412267) - Honest read: These are current-method extrapolations; throughput has been rising fast (better EM, AI segmentation), so "17 My" is a "don't do it this way" number, not a hard floor. But the gap is ~6 orders of magnitude in volume and we still lack a method that scans a whole human brain at synapse resolution at all. This is the central undemonstrated step.
Preservation is the part of the FiO path available today (it doesn't require knowing how to scan or
emulate — only that the information survives until those exist). Deep coverage of the chemistry,
scan-quality, and cryonics debate is in /workspace/cryonics/; this is a summary + 2024–2026 updates.
Claim: ASC — perfuse glutaraldehyde to crosslink/fix proteins, then ramp cryoprotectant and cool below glass transition (~−130 °C) — won the Brain Preservation Foundation Large Mammal Prize (whole pig brain) on March 13, 2018, purse $80,000, lead researcher Robert McIntyre (now "Aurelia Song"), senior author Greg Fahy, at 21st Century Medicine. They had won the Small Mammal Prize (rabbit) in Feb 2016. - Confidence: C1 (two sources) - Sources: BPF Large Mammal announcement; PRWeb release, dated Mar 13 2018 - What was demonstrated: Extensive 3D EM showed the connectome (synaptic ultrastructure) was preserved across the whole pig brain after cooling to and rewarming from below glass transition — "processes were traceable and synapses crisp on FIB-SEM throughout." This is a real, judged, whole-large-mammal-brain structural result. C1. - What was NOT demonstrated: That the preserved information is sufficient to reconstruct the person, or any functional recovery (the tissue is chemically fixed = biologically dead). The prize was explicitly about structural connectome preservation, not viability or "it's still you."
/workspace/cryonics/cryonics-posts/post-1-why-humans-solid/)./workspace/cryonics/alcor-research/). 2026 update: the Fahy/Coles preprint reported
a decade-stored vitrified human brain "astonishingly well preserved," "every detail" visible —
but the connectome was not fully mapped and tissue was fixed after rewarming (so not a viability
proof). See /workspace/cryonics/alcor-research/research/alcor-timeline.md and alcor-writing-catalog.md.| Org | What they do | Status (2026) | Source |
|---|---|---|---|
| Brain Preservation Foundation (Hayworth, pres.) | Ran the prizes; advocacy; defines connectome-at-FIB-SEM-resolution as the bar | Active; prizes won (2016/2018); now advocacy + tech prize | BPF team |
| Nectome (McIntyre/Song) | ASC-based memory preservation; "100% fatal" pitch (perfuse fixative at end of life) | Still operational, no delivered service; $10k refundable waitlist; reportedly ~25 on list incl. Sam Altman; no funding round since 2018; MIT cut ties 2018 | STAT; Tracxn profile; TechCrunch (MIT) |
| Oregon Brain Preservation → "Sparks Brain Preservation" (2025) | Aldehyde-based fluid preservation (fridge temp); offered free research-option preservation (2024) | Active; rebranded 2023 then 2025; arguably the most accessible structural preservation | Sparks history; Oregon Cryo services |
| Cradle → "Until Labs" (2025) (Laura Deming) | Reversible cryopreservation (biological revival), not upload-oriented; recovered electrical activity in rewarmed rodent slice (Feb 2024) | Well-funded ($100M+ total); aims at organ banking → eventually whole organisms | Longevity.tech; Series A → Until |
| Alcor Life Extension Foundation (Scottsdale, AZ) | The incumbent nonprofit: liquid-nitrogen vitrification, whole-body + neuro ("neuropreservation"); the org Hayworth has publicly pressed for connectome evidence | Active since 1972; ~the largest by patients-in-storage; the institution the newer startups define themselves against | Alcor; deep dive in /workspace/cryonics/alcor-research/ |
| Cryonics Institute (Clinton Township, MI) | The other incumbent nonprofit; lower-cost whole-body vitrification | Active since 1976; the budget-tier incumbent | Cryonics Institute |
Note the divergence: Nectome/Sparks/BPF/Hayworth pursue structural preservation aimed at future scan + upload. Cradle/Until and Alcor pursue biological revival (reversible cryo / future medicine). These are different bets about which future technology arrives first — emulation vs. nanomedicine. For FiO, the structural/upload branch is the relevant one. C2.
Beyond the upload-oriented structural-preservation orgs above, there is a distinct, fast-growing
biostasis-startup wing — VC-funded, consumer-facing, reacting against the slow incumbent nonprofits
(Alcor/CI). For the FiO goal this is the layer that determines whether good-quality preservation becomes
accessible at scale before someone needs it.
- Tomorrow Biostasis / Tomorrow.Bio (Berlin) — Europe's first commercial cryonics company (founded
2020 by Dr Emil Kendziorra & Fernando Azevedo Pinheiro), operating a fleet of standby cryo-ambulances
and preserving patients in partnership with the European Biostasis Foundation (EBF) facility in
Switzerland. As of May 2025 it reported 20 humans + 10 pets preserved, >800 members signed up,
and raised a €5M seed (co-led by Blast.Club and Truventuro) to expand into the US (NY/CA/FL). This is
the clearest example of cryonics being run as a funded growth startup rather than a member nonprofit.
C2 · EU-Startups, May 2025,
Tech.eu,
European Biostasis Foundation
- The incumbents they react to: Alcor (1972) and Cryonics Institute (1976) — member-funded
nonprofits, the slow-moving establishment the startups frame themselves against. Cradle/Until (Deming,
above) is the well-funded reversible-cryo startup; Sparks/Oregon is the accessible structural one.
- Cross-link: the chemistry, vitrification debate, Alcor-specific critiques and the 2026 Fahy/Coles
vitrified-human-brain study live in /workspace/cryonics/ (esp. alcor-research/).
- Skeptic take: Tomorrow.Bio's operational improvement (rapid standby/ambulance perfusion) is real and
matters for preservation quality; but like all cryonics it is an unproven bet — no preserved human
(startup or nonprofit) has been revived or scanned-and-emulated, and for FiO specifically the
reversible-revival pitch is the wrong branch (structural/upload is the relevant one, per the divergence
note above). C2/C3.
Claim: The most cited recent technical defense of preservation-for-the-future is McKenzie, Zeleznikow-Johnston, Sparks, … George Church, de Magalhães, "Structural brain preservation: a potential bridge to future medical technologies" (Frontiers in Medical Technology, 2024). - Confidence: C1 - Source: Frontiers 2024 - Its honest stance: distinguishes "verifiable" preservation (demonstrably retains psychological properties) from "experimental" preservation (best-effort, no guarantee); argues the rational move is "make our best effort to determine the necessary structural components of valued information … and attempt to preserve them." Concedes "contemporary brain preservation is unlikely to preserve all psychological states." This is the responsible framing — explicitly hedged. C1.
/workspace/cryonics/alcor-research/research/alcor-news-bad.md).We can image at the needed resolution (synapse-scale, ~nm; EM resolves vesicles and synaptic clefts) — over small volumes. We cannot image a whole human brain at that resolution at any reasonable throughput. The frontier is volume-EM + AI segmentation scaling (fly whole-brain done; mm³ of mouse done). The leap to whole human brain is ~6 orders of magnitude. Resolution is solved-in-principle; throughput and total data volume are not (§2d). C2.
This is the question the 2008 roadmap flagged and that's still open in 2025. Candidate "missing ingredients" beyond the wiring diagram: - Synaptic weights & signs — generally not readable from a static EM scan; estimated from synapse size/count as a proxy (a proxy, not a measurement). C. elegans and the Eon fly both inferred weights from synapse counts rather than measuring them. C2. - Neuromodulation / neuropeptides — diffuse chemical signaling that reconfigures circuits over time/space; not captured by a wiring map. The functional-connectomics literature flags this explicitly. (PMC6630759) C2. - Plasticity / learning rules — a static snapshot is a single frame; it has no dynamics. The Eon fly "cannot form long-term memories." C2. - Glia — astrocytes/microglia modulate synapses and may carry information; usually treated as scaffolding in emulations. Whether they're functionally load-bearing for identity is unknown. C3. - Molecular/sub-synaptic state — phosphorylation states, receptor densities, etc. The roadmap's higher rungs (proteome, single-molecule). Whether any of this matters for you is unknown. C3. - Bottom line: We have no validated answer to "what level of detail preserves the person." Optimists assume the spiking/synaptic level suffices; pessimists point to neuromodulation, plasticity, and molecular state. This is unresolved. Mark anything claiming a specific sufficient level as speculative. C3.
A connectome is structure; emulation needs dynamics + parameters. You can have a perfect map of every wire and still not know the signal each wire carries, how it changes with learning, or how chemicals reconfigure the whole thing. C. elegans (40 years, 302 neurons, complete map, no agreed functional emulation) is the existence proof that mapping ≠ understanding ≠ emulating. C1/C2.
Claim: Eon Systems (senior scientist Philip Shiu; announced March 2026 by Michael Andregg) ran the FlyWire connectome inside a simulated fly body (NeuroMechFly v2 + MuJoCo physics), producing walking, grooming, feeding behaviors at ~91% behavior accuracy — billed as the first embodied whole-brain emulation. Built on Shiu et al.'s Nature 2024 leaky-integrate-and-fire model (~125–140k neurons, ~50 M synapses, ~95% motor-behavior prediction). - Confidence: C1 for the claim's existence; C2/C3 for its significance - Sources: Eon "We've uploaded a fruit fly"; The Register (Mar 16 2026); Mindplex - Honest caveats (stated by Eon itself): weights were "determined by the number of synapses" (inferred proxy, not measured); excitatory/inhibitory assigned via ML neurotransmitter prediction; uses a simple LIF neuron model; no plasticity ("this fly cannot form long-term memories"); motor neurons not actually traced (body wasn't scanned, so they hand-connected to the sim body). - Interpretation (C3): This is a genuine milestone — the full loop (sensing → connectome → motor → embodied action) closed for the first time at whole-brain scale. But it is a generic fly behaving plausibly from a connectome template, not "a specific fly's mind is in there." It validates the method's plausibility at fly scale; it does not demonstrate that you can recover an individual's memories/identity from a scan. The hype-to-substance gap in the popular coverage ("uploaded a fly," "copy-paste consciousness") is large — treat headlines skeptically.
Two distinct questions, both unresolved: 1. Consciousness: Would a functional emulation have experiences? (Hard problem of consciousness.) 2. Identity: Even if conscious, is the upload you, or a copy? Destructive uploading then = you die + a copy lives. (The copy/continuity problem.)
The FiO premise needs a superhuman AI to (a) do the scan-to-model translation at scale (the
data-acquisition + segmentation problem is plausibly AI-tractable in a way human labor is not — AI
segmentation already drives FlyWire/MICrONS), and (b) run and manage the uploaded minds (CelestAI).
- This is a dependency layered on top of the WBE problems, not a substitute. An aligned superhuman
AI could collapse the §2d throughput problem; it does not make the §5b "which details matter?"
question go away (the AI would still need the right data + theory).
- AI timelines and alignment are out of scope here — see /workspace/cryonics/not-die/how-not-to-die-from-agi.md.
Note only: FiO requires both a benevolent superhuman AI and solved WBE; it is a conjunction of
two hard, uncertain things. C3.
Critical path (each must be solved; roughly sequential dependencies): 1. Preserve a brain such that the necessary information survives → partially demonstrated (ASC preserves connectome ultrastructure; "necessary" is unproven). 2. Scan a whole human brain at synapse resolution → not demonstrated (6 OOM beyond current; needs new throughput methods). 3. Know what to extract (which biophysical details are functionally sufficient) → open science problem, no validated answer. 4. Translate scan → running model with correct parameters (weights, signs, dynamics) → not demonstrated at vertebrate scale; even fly uses inferred weights. 5. Run at adequate fidelity → compute is the easy part (exascale exists; fidelity depends on #3). 6. It's conscious & it's you → unresolved philosophy (§6). 7. A superhuman AI to run/manage it (FiO-specific) → external dependency (§7).
Honest probability estimates (my reasoning, not authoritative — treat as calibration anchors, C3): - P(structural brain preservation good enough to preserve a recoverable connectome, available today and used by someone now): ~0.5–0.7 that the structure is preserved; but P(that structure is sufficient to reconstruct the person) is the deep unknown — call it 0.15–0.5, genuinely uncertain. Reasoning: ASC's structural result is real and judged; sufficiency depends on §5b, which is open. - P(whole human brain scanned at synapse resolution by 2050, given no transformative AI): low, ~0.1–0.25. Reasoning: 6 OOM throughput gap; historically connectomics scaled ~10×/several years. - P(same, given a transformative/superhuman AI by then): much higher, because AI plausibly cracks throughput+segmentation — but this just relocates the bet onto AI timelines (out of scope). - P(a functional human WBE — actually reproducing a person's cognition — by 2050): low, ~0.05–0.15. Reasoning: gated by §5b ("what to extract") which is an unsolved science problem, not just engineering. - P(full FiO: aligned superhuman AI and solved WBE and identity-preserving upload): very low this century absent a singularity-class event; a conjunction of several <0.3 terms. I'd put it single-digit % by 2100 conditional on no civilization-ending events — dominated by the "what-to-extract" and "is-it-me" unknowns, not by compute. - Most leveraged near-term action for an individual targeting FiO: ensure good structural preservation (it's the only rung available now and it's the prerequisite for all later rungs). The emulation/AI rungs are not actionable by an individual today. C3.
Net: The roadmap was right that this is a well-posed problem reachable by extrapolation — but the extrapolation is enormous, the binding constraints are data-acquisition and an unsolved science question ("what level of detail is you?"), and the philosophy is unresolved. We have moved from "0 animals emulated" to "1 fly behaving from its connectome" in ~40 years. That is real progress and real distance.
Maturity legend used above: Demonstrated = judged/published result; Partially demonstrated = real result that doesn't establish the FiO-relevant claim; Not demonstrated = no existing method; Open science problem = nobody knows the answer; Philosophy = not empirically settled.