MEV and Restaking in 2026: What Actually Happened

TL;DR

No. There is no production encrypted mempool on Ethereum mainnet. The flagship attempt — Shutter and Primev's threshold-encrypted mempool for the proposer-builder separation supply chain, announced on 2 December 2025 with mainnet launch targeted for December 2025 or January 2026 — has not shipped. On 14 September 2026, Primev's own documentation states the preconfirmation and mev-commit stack is "currently in maintenance mode", the settlement explorer at mev-commit.xyz is decommissioned, and the surviving Validator Coalition is explicitly "not extra block revenue from a live commitment market". Shutter's threshold encryption is live only on Gnosis Chain, with a 7-keyper alpha set. The protocol-level alternative, EIP-8105, is a December 2025 draft that does not appear in the Ethereum Foundation's September 2026 tier list of 62 Hegotá EIP candidates. And block building did not decentralise: Titan built 55.78% of blocks over 7 days and 62.87% over 24 hours, with BuilderNet at 10.15%. Restaking, separately, is down 70% from peak — EigenLayer $22.06B to $6.59B — and slashing went live on 17 April 2025, not 2026.

Is There an Encrypted Mempool on Ethereum? The Status of Each Attempt

The short answer is above. Here is the state of each line of work, with dates, because the search results on this question are full of announcements presented as shipped products.

Shutter and Primev, PBS-layer threshold encryption. Announced on 2 December 2025: the two would bring the first threshold-encrypted mempool to Ethereum's PBS supply chain, whose published roadmap listed the proof of concept as "Coming soon" and "December 2025/Jan 2026: Mainnet launch". Checking Primev's documentation↗ on 14 September 2026: the preconfirmation and mev-commit stack is described as "currently in maintenance mode". The docs say "The old dashboard is deprecated". The mev-commit.xyz domain no longer serves a settlement explorer at all: it now returns a 308 redirect chain ending at primev.xyz/ai, Primev's AI product page. What survives is the Validator Coalition, which the docs are careful to say is "not extra block revenue from a live commitment market". That is a project in wind-down language, not launch language.

Shutter on Gnosis Chain. Threshold encryption is genuinely running, on the Shutterized Beacon Chain, with a 7-keyper alpha set. That is a real deployment and a real result. It is not Ethereum mainnet and the keyper set is small enough that you should reason about it as an alpha, which is what it is called.

Preconfirmations. The other big 2025 promise in this area. The commitment market that was supposed to carry them is the same stack now in maintenance mode.

EIP-8105, enshrined at the protocol level. A December 2025 draft. Covered in the next section.

If you are choosing a mempool privacy strategy today, none of the above is the answer. The answer is still private order flow: send the transaction to a builder or a protected RPC rather than to the public mempool, and accept that you are trusting whoever you sent it to. I go through the application-side defences in front-running and MEV protection. Nothing in 2026 changed the underlying shape of that problem.

EIP-8105: The Enshrined Encrypted Mempool Draft

This is the most interesting unshipped thing in the area, and almost nobody has written it up.

EIP-8105, "Universal Enshrined Encrypted Mempool", was first drafted on 17 December 2025↗ as GitHub PR #10943. The design splits a transaction in two:

• A visible envelope carrying gas parameters and fees, so builders can still order and price the transaction.
• An encrypted payload carrying calldata, recipient and value, decrypted only once block ordering is fixed.

The key-provider registry is deliberately technology-agnostic: threshold encryption, MPC, TEEs or FHE can all sit behind it. That is the right call for something meant to last more than one cryptographic fashion cycle.

Two numbers and one honest tradeoff from the draft:

That last row is the part to internalise. An enshrined encrypted mempool stops a builder from front-running you on the way in. It does not make your transaction private. If your threat model is an analyst reading the chain a week later, this EIP does nothing for you, and the draft says so itself.

EIP-8105 does not appear in the Ethereum Foundation's Hegotá EIP evaluation↗ published on 7 September 2026, which graded 62 candidates. I should be precise about what that means: the absence is my reading of the published tier-list summary, not an explicit rejection. The EF graded 2 S-tier, 15 A-tier, 8 B-tier, 7 C-tier, 28 declined-for-inclusion and 2 TBD. EIP-8105 is not among the listed items. Treat it as "not on the near-term roadmap", not as "killed".

Block Building Did Not Decentralise: Titan at 56–63%

The story everyone tells is that BuilderNet, TEE-based building and PBS spread block production out. The measurements say one builder dominates.

From Relayscan↗, captured at 15:04:23 UTC on 14 September 2026, at slot 15,214,520:

Over the trailing 24 hours Titan was higher still: 62.87%, 4,157 blocks. Builder profit over the 7-day window totalled 821.5 ETH, of which Titan took 336.87 ETH.

I ran my own 120-block sample the same afternoon as a cross-check and it matched the shape: Titan 52%, Quasar 19%, BuilderNet 10%, Eureka 6%, bombora 5%. Beaverbuild and rsync — two names that appear in almost every MEV explainer written in the last three years — did not appear in my sample at all.

Two caveats I will not bury. Builder share moves, and a 24-hour window and a 7-day window on the same snapshot differ by seven percentage points, so quote the window with the number. And Relayscan warns that its profit figures are derived from block.coinbase balance deltas, which miss off-chain builder payments — so 821.5 ETH is a lower bound on builder revenue, not a measurement of it.

The conclusion survives both caveats. One builder producing the majority of Ethereum blocks over a full week is a concentrated market by any reading. I am not going to call it more concentrated than the pre-BuilderNet period without a dated figure from that period next to it, and I do not have one I can source. BuilderNet holding 10.15% is not the outcome its launch coverage described.

The Relay Layer Is More Concentrated Than the Builder Layer

The builder number gets quoted occasionally. The relay number almost never does, and it is worse.

Relay shares over the same 7-day window on 14 September 2026:

Those three account for 88.3% of all relay payload deliveries. Relayscan logged 96,428 deliveries against 45,938 blocks over the same window, because mev-boost requests the payload from every relay that returned the winning bid, so read this as a share of deliveries rather than a share of distinct blocks. One of them is named "regulated", which is the whole censorship-resistance conversation in a single subdomain. Relays sit between builders and proposers, so they are a chokepoint for both inclusion and ordering, and there are effectively three of them that matter.

This is the part of the MEV supply chain where a developer's exposure is structural rather than transactional. You can route around a builder. You cannot route around the fact that three relays carry the block flow.

What PBS Delivered, and What Slipped to December 2026

Proposer-builder separation as currently deployed is out-of-protocol: it runs on relays and mev-boost, not on consensus rules. Enshrining it is EIP-7732, and it is a headliner for the Glamsterdam upgrade along with EIP-7928, block-level access lists.

Glamsterdam has slipped. The Ethereum Foundation's February 2026 priorities post scheduled it for H1 2026. The Foundation's separate 7 September 2026 priorities post↗ now works from "Shipping Glamsterdam in December 2026", with the Hegotá fork following and client teams able to start Hegotá implementation in late Q4 2026. That is a genuine schedule slip, stated by the same source in two posts, and worth quoting both dates rather than either one.

The testing is real and public. Platåberget↗ was announced on 17 August 2026 and forked on 20 August 2026, ahead of Sepolia and Hoodi. The Foundation calls it "a short-term testnet" but adds that "Unlike the short-lived devnets before it, Platåberget is intended to run for a few months". If you maintain infrastructure that touches block construction, that is where the ePBS changes are testable now.

So the honest position on PBS: the out-of-protocol version is what runs mainnet and it produced the concentration measured above; the in-protocol version is on a testnet and scheduled for December 2026, roughly six months later than it was scheduled in February 2026.

FOCIL Is the Censorship-Resistance Answer That Got Graded S-Tier

Here is the thing the encrypted-mempool conversation keeps missing. The Ethereum Foundation graded 62 EIP candidates for Hegotá and named exactly two must-ship items. One of them is EIP-7805, FOCIL — fork-choice enforced inclusion lists — the consensus-layer headliner for censorship resistance. The other is EIP-8141, Frame Transactions, the execution-layer headliner and the Foundation's chosen path for native account abstraction and post-quantum signatures.

Two S-tier items out of 62 candidates. That is the clearest signal available about where protocol effort is going, and it says the censorship-resistance answer the EF is committing to is enforced inclusion, not encrypted ordering.

The distinction matters for what you can promise a user. Inclusion lists attack the question "can my transaction be kept out of a block?" An encrypted mempool attacks the question "can someone see and reorder my transaction before it lands?" They are different problems. Only the first one has a must-ship EIP behind it.

Also worth noting from the same tier list: EIP-8182, private ETH and ERC-20 transfers, was declined for inclusion in favour of Frames-based alternatives. Privacy at the protocol level in 2026 keeps losing to designs that route through the account abstraction path instead. The account-level stack is where the practical wins are, which is what I measured in account abstraction, measured on-chain.

Restaking: the Real Curve, Not the Retrospectives

Now the second half, where the published numbers are worse than merely stale — several are wrong.

EigenLayer, from the DefiLlama protocols API on 14 September 2026 (the exact call is in the reproduction section below):

That is -48% year to date in 2026 and -70% from the August 2025 peak, with a partial recovery off the 1 July 2026 low of $4.29B.

The rest of the category is worse:

• Symbiotic: peak $2.70B on 7 December 2024, now $0.44B. Down 84%.
• Karak, now listed as OpenGDP Shared Security: peak $1.05B on 7 June 2024, now $0.008B. Down 99%.
• Restaking category total: $10.09B on DefiLlama, of which $3.21B is Babylon, a Bitcoin staking protocol. Symbiotic is filed under Collateral Markets and is not inside that number.
• Liquid restaking: $1.38B, of which Kelp is $1.079B — 78% of the entire liquid restaking sector in one protocol. Renzo has fallen to $0.118B.

A sector where one protocol is 78% of a sub-category worth $1.38B is not a sector with a competitive market in it.

The Slashing Date Half the Internet Gets Wrong

This one is checkable in thirty seconds and it is wrong in prominent places.

EigenLayer slashing went live on 17 April 2025. The EigenCloud announcement↗ tweet, status ID 1912885738116620670, decodes via Twitter's snowflake ID scheme to 2025-04-17T15:07:54Z. The pre-announcement, status 1907525899064660112, decodes to 2 April 2025.

Articles dated April 2026 claim slashing activated in April 2026 and put restaked ETH somewhere between fifteen and twenty billion dollars at that moment. I am characterising them rather than quoting a specific piece, so weigh that accordingly — the checkable half is the series, and DefiLlama has EigenLayer at $9.92B on 17 April 2026. DefiLlama has EigenLayer at $9.92B on 17 April 2026 and $6.59B now. Both the year and the magnitude in those pieces are wrong, by a full year and by two to three times.

This is worth dwelling on as a research habit rather than as a gotcha. A large share of top-ranking "2026" Web3 articles are generated content that backdates 2025 events by twelve months and inflates the accompanying figures. The tell is a round-numbered range with no dated source next to it. Decode the timestamp, check the series, then quote. It took me two lookups to falsify a claim that is currently ranking.

Denominate in ETH Before You Call Anything a Collapse

Most 2026 dollar declines are partly ETH price. ETH went from $4,607 on 15 September 2025 to $2,477 on 14 September 2026, down 46%. Any dollar-denominated drawdown measured over that window inherits most of that.

Run EigenLayer's 2026 numbers through it. On 1 January 2026, $12.60B at an ETH price of $2,967 is about 4.25 million ETH. On 14 September 2026, $6.59B at $2,477 is about 2.66 million ETH. So the year-to-date fall is -48% in dollars and about -37% in ETH terms.

That is still a real contraction — a third of the deposits left in ETH terms, not just in price — but it is not the -48% the headline number implies. The same correction applies to the whole DeFi picture: total TVL fell 23% in 2026 from $114.4B to $88.6B while ETH fell 17% over the same window, and in ETH terms the same TVL fell only about 7%, so roughly two thirds of that particular decline is price and one third is capital actually leaving.

If you are writing about this category, do the ETH conversion before you reach for the word collapse. If you are reading about it and the author did not, that tells you something about the rest of the piece.

What Is an AVS Actually Worth Paying For Once Slashing Is Real?

This is the engineering question the coverage never gets to, because the coverage is still selling the primitive.

Before slashing, an actively validated service bought you a promise with no enforcement behind it. After 17 April 2025, the promise has teeth: misbehaviour can cost the operator real stake. That changes the question from "how much stake is behind this?" to "what specific failure does the slashing condition actually punish, and is that the failure I am worried about?"

I could not find a reliable public figure for 2026 AVS fee revenue, so I will not invent one, and I would treat any article that quotes one without a dated primary source as unverified. What I can give you is the framework I use when a client asks whether an AVS belongs in a design.

Three questions, in order:

1. What is the slashing condition, in one sentence, and does it map to your actual failure mode? "The operator signed two conflicting attestations" is enforceable. "The operator provided bad data" usually is not, because provability is the whole problem. If the condition does not cover your failure mode, the stake behind it is decoration.
2. Is the slashable stake larger than the value at risk in your application? If you are securing $50M of settlement with an operator set whose slashable exposure is $2M, the economic security argument does not close. This is an arithmetic check, not a judgment call, and most designs fail it.
3. Would a trusted committee with a written SLA give you the same guarantee more cheaply? Often yes. Restaking buys you cryptoeconomic enforcement against a permissionless operator set. If your operator set is five known companies anyway, you are paying protocol complexity for a property you are not using.

The honest default in 2026: an AVS earns its place when you need a permissionless operator set whose misbehaviour is objectively provable on-chain. Data availability attestation and oracle-style signing fit. Most application logic does not. A staking design where the reward and penalty mechanics have to be right is its own engineering problem — I work through the mechanics in staking protocol reward distribution.

The capital flight in the table above is not irrational. It is what happens when a category is repriced from "yield on ETH with no enforcement" to "yield on ETH with enforceable downside", and a lot of the deposits turn out to have been there for the first thing.

How to Reproduce These Measurements

None of this needs a paid data provider. The whole point is that you can check it.

# Builder and relay share, 24h and 7d windows
open "https://www.relayscan.io/overview?t=7d"

# Restaking TVL series, dated points
curl -s "https://api.llama.fi/protocols" \
  | jq '.[] | select(.name|test("Eigen|Symbiotic|Karak|Kelp|Renzo")) | {name, tvl}'

# Decode a tweet timestamp from its status ID (snowflake epoch 1288834974657)
python3 -c "import datetime;i=1912885738116620670;print(datetime.datetime.utcfromtimestamp(((i>>22)+1288834974657)/1000))"

For builder share, sampling blocks yourself is a useful cross-check against a dashboard: pull the extraData field or the fee-recipient address across a few hundred recent blocks and count. My 120-block sample is about 24 minutes of chain. At a 55.78% true share the standard error on 120 draws is roughly 4.5 percentage points, so landing within four points of Relayscan is inside one standard deviation and confirms nothing about the dashboard. Treat it as a check that the ordering of builders is what the dashboard says, not as independent verification of the shares.

One resource that no longer works, and is still cited as live in current tutorials: eigenphi.io, the most-referenced public MEV analytics dashboard of the 2023-24 cycle, now returns a 301 redirect to a Substack. The dashboard is gone. That is also why I do not quote a total-MEV-extracted figure anywhere in this article — I could not find a reliable 2026 one, and the 821.5 ETH of 7-day builder profit from Relayscan is a lower-bound proxy, not an MEV total.

If you are building on top of any of this, the same discipline applies to the cost side of the stack, which I measured in Fusaka, blobs and rollup economics, and to the chain risk underneath it in what Stage 1 and Stage 2 actually mean. Solvers in particular carry MEV exposure on both their fills and their rebalancing legs — I cover that in building an ERC-7683 solver.

Key Takeaways

• No production encrypted mempool exists on Ethereum mainnet. Primev's documentation described the mev-commit stack as "currently in maintenance mode" on 14 September 2026, with the settlement explorer decommissioned; Shutter's threshold encryption runs only on Gnosis Chain with a 7-keyper alpha set.
• EIP-8105 buys ordering confidentiality, not privacy. The December 2025 draft states "<1 slot overhead" for encrypted transactions and says plainly that transactions are published after inclusion. It is not in the EF's 62-EIP Hegotá tier list.
• Block building got more concentrated. Titan built 55.78% of blocks over 7 days and 62.87% over 24 hours on 14 September 2026; BuilderNet held 10.15%. Three relays took 88.3% of relay payload deliveries.
• The EF's censorship-resistance bet is FOCIL, not encryption. EIP-7805 is one of only two S-tier items out of 62 Hegotá candidates; the other is EIP-8141, Frame Transactions.
• Glamsterdam slipped from H1 2026 to December 2026, per the Foundation's own February and September 2026 posts. The Platåberget public testnet forked on 20 August 2026.
• EigenLayer slashing went live on 17 April 2025, not 2026. The announcement tweet's snowflake ID decodes to 2025-04-17T15:07:54Z; articles claiming April 2026 with fifteen to twenty billion dollars restaked are wrong on both the year and the number.
• Restaking is down 70% from peak — $22.06B to $6.59B — with Symbiotic down 84% and Karak down 99%. In ETH terms the 2026 fall is about -37%, not -48%.

About the Author

I'm Uvin Vindula — a Web3 and AI engineer based between Sri Lanka and the UK. I build and audit production smart contracts, and I check protocol claims against the chain and the primary announcement before repeating them, which is how the slashing date in this piece got corrected. You can see my work at iamuvin.com or reach out about a project at hello@iamuvin.com↗.

If you are deciding whether restaking, an AVS or a private order-flow route belongs in your architecture, let's talk about your project.