Restaking was a genuinely novel cryptoeconomic primitive when it launched. Letting the same staked ETH simultaneously secure multiple Active Validated Services produced real composability gains and gave AVSs a way to bootstrap security without inventing new tokens. At small scale, the additional risk was bounded and the benefits were clean. At the scale restaking has now reached — over $20 billion in restaked ETH across EigenLayer and its imitators — the math has changed in ways the industry has been slow to articulate.
The core problem is correlation. Restaked ETH is, by definition, simultaneously underwriting the cryptoeconomic security of multiple unrelated services with heterogeneous risk profiles. A bug or coordinated misbehavior incident in any one of them can cascade slashings into the same collateral that secures the base layer — exactly the property that the original thesis claimed to be safe. The math is fine when the AVSs are conservatively scoped and the slashing conditions are tight. Both of those qualifiers are now actively eroding under competitive pressure, and the industry has not yet built the language to discuss the erosion honestly.
The mechanics of the erosion are predictable. AVSs that pay the highest yield attract the most restaked stake. Highest-yield AVSs are, structurally, the ones with weakest slashing conditions or most aggressive promised cash flows. Restakers, optimizing for yield, allocate disproportionately into exactly those services. Risk concentrates precisely where the protocol's defenders argue it doesn't. Recent dashboards from Gauntlet and Chaos Labs show that the top-yielding AVSs on EigenLayer command roughly 60% of restaked stake, despite representing less than 15% of services by count. That is not a balanced exposure profile — it is a yield-chasing one. Renzo and Kelp DAO, the two largest liquid-restaking protocols, both publish AVS-allocation breakdowns; the cohort weights are even more concentrated than the headline EigenLayer numbers suggest, with the top three AVSs by yield commanding nearly 75% of restaked exposure within those products. Anyone who tells you the design is risk-diversified should be asked to explain those numbers, in writing, to a regulator.
The defenders' arguments are not wrong, just insufficient. AVS slashing conditions are individually scoped, and on-chain insurance markets like Nexus Mutual have begun pricing residual risk. Protocol-level guards distinguish "consensus-critical" slashing from "unilateral" slashing, with the former kept rare and the latter capped per-AVS. EigenLayer's veto-committee mechanism adds an additional human-supervised circuit-breaker. All true. None of it changes the fundamental fact that under stress, a multi-AVS slashing cascade routes losses into a single, overlapping pool of collateral. Insurance markets cannot diversify a risk that is, by construction, undiversifiable at the layer they cover.
The honest counterargument is that the same logic applies to traditional banking and reinsurance — both rely on overlapping capital pools that, in extreme tail events, do correlate. We accept those risks because the systems have evolved supervisory frameworks, lender-of-last-resort backstops, and regulatory capital requirements that smooth out tail outcomes. Restaking has none of those backstops. The industry's response — that on-chain governance and insurance-market pricing will substitute for those institutional layers — is a hypothesis, not a track record. The next stress event will be the test, and stress events tend to arrive on schedules nobody predicted.
The 2026-2027 question is not whether a stress event happens; it is when, and how systemic the cascade ends up being. The best near-term outcomes for the category involve a small, contained slashing incident on a low-stake AVS that exposes the cascade dynamics without destabilizing the base layer — essentially a fire drill. The worst outcomes involve a coordinated incident across two or more AVSs that triggers cascading insurance claims, forced unstaking flows, and a reflexive ETH price drawdown that strains the entire restaking edifice. The category does not have to die. But the industry should stop pretending the design is risk-free and start pricing the structural correlation honestly. The first step is admitting the problem out loud.