Hook
Over the past disclosure window, a single enterprise remote-access gateway lost the same class of boundary for the third time. SonicWall's SMA1000 line โ the SSL-VPN appliance thousands of firms park at the edge of their networks โ shipped another pre-authentication server-side request forgery (CVE-2026-102255), rated CVSS 10.0, with no mitigation path beyond a firmware upgrade. The first cycle read as an accident. The second read as a pattern. The third reads as a decision. And that decision โ letting an edge proxy act on behalf of a requestor before verifying who the requestor is โ is the same decision crypto's bridges, oracles, and intent solvers make every single day. The confused deputy is not a vulnerability. It is an architecture. I see the pattern before it becomes a trend, and this one has been trending across enterprise security and on-chain infrastructure in parallel for years. The company names rotate. The CVE numbers climb. The structural error stays fixed in place, like a crack that outlives every contractor hired to paint over it.
Context
The formal name for what happened is CWE-441, the confused deputy. A component with legitimate authority โ here, an edge reverse proxy at the Work Place portal, the /wsproxy path, and the Appliance Management Console โ is tricked into performing an action using its privileges on behalf of someone who never proved they deserved them. The device isn't broken the way a bad line of code is broken. It is broken the way a building is broken when the front desk hands out master keys before checking IDs. Authentication is placed after the trust decision, not before it.
That inversion is what makes the SMA1000 case worth more than a patch note. Across three disclosure cycles, the opening move never changed: an edge pre-auth SSRF (CVE-2026-15409, then CVE-2026-83548, then CVE-2026-102255). Behind it, each time, sat an authenticated execution chain โ command injection, path traversal, stored XSS โ so the pre-auth hole was never the payload; it was the doorway. When the same entrance appears three times, it is not a coincidence of bugs. It is a record of design. The patches were real. The fix was not.
Now translate the mechanics. A cross-chain bridge is a confused deputy with a treasury. It holds authority over assets on two chains and acts on messages it did not itself originate. If its validation layer trusts a relayer's assertion before independently verifying it, the bridge will happily move funds on behalf of an attacker who merely sounded like the other chain. That is not a coding error. That is a trust boundary drawn in the wrong place โ exactly the SMA1000 error, denominated in ETH instead of dollars.
Core
Here is where my audit background sharpens the reading. In 2017, during the ICO mania, I spent six months manually reviewing forty-plus ERC-20 contracts for a mid-tier payment token. The vulnerability I found was not clever โ a reentrancy flaw in the distribution logic, worth roughly $2.5 million โ and it lived precisely where the contract trusted a callback before checking its own state. The code compiled. The tests passed. The design trusted too soon. Every bridge exploit since has rhymed with it.
Look at the recurring anatomy. Nomad lost roughly $190 million in 2022 because an initialization routine left one trusted root at a value anyone could overwrite โ a bridge that accepted a message it never authenticated. Wormhole lost about $326 million the same year when a signature-verification path was bypassed, letting the guardian layer vouch for a mint that never happened. Ronin, roughly $625 million, fell because a threshold of validator keys sat in one trust cluster that could be socially engineered. None of these were cryptographic breaks. They were boundary breaks โ the deputy trusted the request before the request earned it.
The oracle layer repeats the pattern in slower motion. Oracle feed latency is DeFi's Achilles' heel, and a curated node set solving decentralization is a solution wearing the costume of the problem it claims to fix. In 2020, I modeled impermanent-loss dynamics on a USDT/ETH pair for three weeks and watched algorithmic stablecoins quietly redistribute wealth from retail to whales. The mechanism was not malice. It was a price feed trusted a beat too early, and a pool that acted on it. Latency is just the confused deputy's clock. That is why a twenty-second stale feed inside a liquidation cascade becomes a nine-figure event rather than a rounding error.
Intent-based architectures are the newest costume. A user signs an intent; a solver network competes to fulfill it. The pitch is that the user never touches a mempool, never fights MEV. But the MEV didn't vanish โ it relocated. The solver becomes the new deputy, holding discretionary authority over execution, and the attack surface moves from on-chain transparency into off-chain solver logic where nobody can watch. Intent systems won't replace DEXs; they will simply move the manipulation from a public mempool to a private solver network, and call the move an improvement. We map the flows, but the ocean remains unmapped.
And note the surface design. The SMA1000 exposed three management planes โ portal, /wsproxy, and console โ each carrying its own independent flaw, which tells you the vendor never converged its attack surface under one governance model. Crypto's version of that failure is the protocol with six deployed chains and six slightly different trust assumptions, none of them reconciled.
Contrarian
The reflexive response to a third SMA1000 disclosure is to blame the vendor's engineering. The reflexive response to a bridge hack is to blame the auditors. Both miss the actual cause. The cause is economic: architecture-level repair is expensive and slow, while patching is cheap and visible. A mature hardware line has every incentive to keep shipping point fixes to a product it no longer wants to rebuild โ and a mature crypto protocol has every incentive to keep shipping audits and bug bounties rather than re-drawing the boundary that keeps failing. The industry treats the symptom list as the disease, because the symptom list is what gets funded. Reputation is a lagging indicator of architecture, and architecture is what nobody is paid to rebuild.
The contrarian reading is darker. Three disclosure cycles don't prove SonicWall is uniquely negligent โ Cisco and Splunk ran parallel edge-device flaws in the same window, which means this is a category-level condition, not a brand-level sin. That's cold comfort. "Everyone has the same problem" is simultaneously a valid defense and the most precise description of a systemic failure. DeFi promised freedom; it delivered a mirror. The mirror reflects the same debt the enterprise world has been carrying: trust granted before verification, because verification is slower and trust is cheaper โ until it isn't.
Takeaway
So watch for the fourth cycle. Not because I want it, but because the design debt hasn't been repaid, and unpaid debt compounds. The signal to track is not whether the next patch ships โ it will โ but whether the boundary moves: authentication before authority, least privilege for the deputy, isolation between edge and core. Until that happens, in Lagos or in a validator set, the same doorway stays open. Between the wire and the wallet, there is a void โ and someone is always standing in it, waiting to be trusted.