Three disclosure cycles. One entry point. A CVSS 10.0 that lands before credentials do.
SonicWall's SMA1000 remote-access gateway just shipped its third pre-authentication SSRF, disclosed under advisory SNWLID-2026-0017. No workaround. No mitigation. Patch or replace. The advisory reads like a rerun: the Work Place portal, the /wsproxy edge, the appliance acting on behalf of a requester it has not yet verified. The CVE numbers climb โ 15409, then 83548, then 102255 โ but the shape does not move. Same edge. Same pre-auth. Same confused deputy.
I have seen this pattern before. Not on a firewall. On bridges. On RPC endpoints. On every edge layer crypto built and then forgot to authenticate.
The SMA1000 is a mature SSL-VPN appliance โ hardware 6210 and 7210, a virtualized 8200v. It lives at the network perimeter. Its job is to be a reverse proxy: terminate sessions, forward requests inward, represent the client to internal services. Three surfaces face the world โ the Work Place portal, the /wsproxy handler, and the Appliance Management Console.
What makes those three surfaces matter is that they are not one door. They are three doors into the same house, each with its own lock, and all three have now been opened at least once. Work Place, /wsproxy, and the AMC were never converged behind a single authentication choke point. Crypto has the identical disease. A public RPC, an indexer API, a bridge relayer, a governance forum, a Discord bot holding a hot key โ each one is an edge, and each one quietly assumes the layer behind it will catch whatever it lets through. None of them do.
Here is the flaw. In the pre-authentication state, the edge proxy acts on the requester's behalf before it has verified who the requester is. That is a confused deputy (CWE-441). The proxy holds authority the caller does not have, and it spends that authority for the caller. The trust decision happens after the privileged action, not before it.
Now map it onto crypto, because the architecture is identical.
Every RPC endpoint is a confused deputy. You send a JSON-RPC call. The node relays it, signs nothing, and acts with the node's own privileges โ peer connections, mempool access, state reads. The caller proved nothing. The bridge relayer is a confused deputy: it holds mint authority or validator weight and executes on a message the edge never cryptographically verified. The intent solver is a confused deputy: it takes your signed intent and fills it with its own capital and routing, and you never see the path your order took. Authority lives at the edge. Authentication lives further in โ or nowhere at all. The yield was real; the trust was phantom.
The single most important signal in the SonicWall disclosure is not the CVSS score. It is the repetition. Three disclosure cycles, three times the same opening move โ an edge pre-auth SSRF. When the same entry point appears three times, you are not looking at vulnerability coincidence. You are looking at a design record. A patch fixes a symptom. It does not move a trust boundary. The boundary is wrong at the architecture level: the proxy layer carries too much privilege, and authentication sits after the trust judgment instead of before it.
I have watched crypto make the same mistake, and I have made it myself. In 2025 I shipped an AI-driven portfolio rebalancer. Based on my audit experience, the first thing I checked was not the model. It was the authority graph. The agent could move funds before it had verified its inputs. My own code was the confused deputy โ the exact CWE the SonicWall advisory names, wearing a neural net instead of a VPN. I fixed it by moving verification in front of the privileged call. That is the whole fix. It is also the fix almost nobody ships, because it costs an architecture, not a patch.
The depth of the SonicWall failure is what should terrify you. The pre-auth SSRF is only the door. Behind it sit post-auth command injection (CWE-78), a Zip Slip path traversal (CWE-22), and stored XSS (CWE-79). Pre-auth entry, then post-auth execution. That is not four bugs. That is one kill chain with four links. Defense-in-depth failed at every layer at once, which tells you the layers were never independent โ they all inherited the same wrong trust boundary.
Crypto even has its own version of the Appliance Management Console. Restaking layers promise that slashing will punish misbehavior, but slashing is a post-hoc trust judgment โ it fires after the privileged action, not before it. A light client that verifies headers lazily is the same architecture: it trusts the edge of the chain and reconciles later. Every one of these designs puts the cart of execution in front of the horse of verification, and every one of them is a confused deputy wearing a different hat.
Crypto's kill chains inherit the same debt. An unauthenticated RPC call routes a transaction; the transaction hits a contract with an unchecked privileged function; the privileged function mints or drains. Nomad. Wormhole. Ronin. Different chains, same skeleton: the edge trusted a message it had not proven. Chaos is just a pattern waiting for a label.
Then there is the economics, and this is where crypto is worse, not better. A mature hardware line can at least push firmware 12.4.3-03670 and 12.5.0-03082. A deployed smart contract cannot be patched at all. You cannot move a trust boundary on-chain โ you can only migrate, and migration means TVL flight. So teams keep the broken boundary and bolt on monitoring. Hope is a terrible hedge against a black swan. A firewall at least has a vendor on the hook. A bridge has a multisig and a prayer.
The history sharpens the picture. This is a product line that has already seen MFA seeds stolen and ransomware actors โ tracked as UTA0533 โ operating against it. A pre-auth CVSS 10.0 with no mitigation is exactly the profile that lands a CVE in the CISA KEV catalog, which converts a vendor problem into a federal deadline. Once that happens, every downstream customer inherits a clock they did not set. In crypto there is no KEV. There is no catalog, no deadline, no authority that can force a fix. The vulnerability just sits there, and the TVL just sits there, until someone drains it.
The comfortable read of this advisory is industry-wide problem. SonicWall's defenders will point to Cisco FMC and NX-OS shipping the same class of pre-auth bug. The crypto crowd has its own version: all bridges get hacked, it is the cost of doing business. Both are cop-outs, and they do the same damage. Calling a structural failure industry-wide is an immunity narrative. It normalizes the failure and kills the urgency to refactor. The moment you file a design debt under everyone has it, you have decided to keep paying interest forever.
The deeper contrarian point is this: decentralization makes the confused deputy worse, not better. A centralized appliance can force a patch window. A decentralized protocol cannot force-migrate four hundred validators who are economically comfortable. So the trust-boundary debt that SonicWall can eventually retire is the debt crypto structurally cannot. The code-is-law crowd should be the most frightened people in the room โ because when the law is a contract, the only remedy is a fork, and forks are the most expensive patch in existence.
Watch the edges. Watch the pre-auth. The next CVSS 10.0 will not arrive with a firewall logo on it. It will arrive as an RPC endpoint, a relayer, a solver โ and it will be dressed as innovation. We traded sleep for alpha, and alpha for scars. For the traders reading this: check what your venue's edge actually trusts. For the builders: move the auth in front of the call. For everyone else, the pattern is the warning. The question for the next twelve months is not whether a fourth disclosure cycle lands. It is whether the industry refactors the boundary before it does, or waits, like a mature hardware line, until the customers leave.