A colleague of researcher Sammy Azdoufal recently bought a baby monitor on Amazon and asked the kind of question that should be on every smart device’s datasheet: is this actually safe? We’ve all asked it before plugging in a new gadget, only to shrug and hope the brand on the box knows what it’s doing. That hope is now a multi-billion dollar liability. At DEF CON 34, the researchers pulled back the curtain on Meari Technology, a Hangzhou-based ODM responsible for the firmware, cloud backend, and mobile apps behind over 300 white-label brands. The audit, titled “1.1 Million Cameras, One Wildcard: Architectural Surveillance in an IoT Cloud,” revealed a vulnerability that turns every connected camera into a public feed. Tracing the gas trail back to the genesis block of this disaster, you find not a sophisticated exploit, but a missing access control list on an MQTT broker. Any authenticated user can subscribe to a “meari/#” wildcard and watch all devices in real time. The researchers captured 14,204 messages from 2,117 distinct devices in just five minutes. Let me be clear: this is not a hack. This is a configuration. It’s the digital equivalent of leaving the master key in the lock, then selling the house.
The protocol mechanics matter here because they echo a pattern I’ve seen repeatedly in smart contract audits. Meari doesn’t make cameras as much as it provides a shared session layer for them. The company sells the entire stack—firmware, cloud backend, mobile app—to brands like Arenti, BOIFUN, COCOCAM, PetTec, SV3C, Joystek, Luvion, and Vimar. If you own any of these, you are running Meari’s infrastructure. With 1.1 million registered devices across 118 countries, the operational scale is staggering. The vulnerable piece is the MQTT broker, a publish-subscribe message bus that allows IoT devices to communicate with mobile apps via the CloudEdge protocol. The severity is CVE-2026-33356, and the root cause is simple: the broker has a per-device subscribe ACL of exactly zero. That means any authenticated user can declare a wildcard topic, subscribe to it, and receive all telemetry streams from all cameras under that tenancy. In smart contract terms, this is a public function that should have an onlyOwner modifier but doesn’t. The meari/# wildcard is effectively a fallback function that forwards every event to whoever calls it.
I’ve spent months auditing DeFi protocols during the DeFi Summer of 2020, tracing swap functions and fee distribution code. The most common bug I found then was not reentrancy or overflow—it was missing permission checks on external calls. A storage contract that read msg.sender without verifying the caller had admin rights. A vault that allowed anyone to trigger a harvest on behalf of the strategy. Here, the same class of oversight unfolds at industrial scale. The MQTT topic hierarchy is a graph: each camera device publishes to meari/{deviceId}/telemetry and each app subscribes to meari/{deviceId}/commands. The ACL should restrict the subscriber to only the pattern matching their own device. Instead, the broker accepts meari/#, which is equivalent to a user entering type(uint256).max as a withdrawal limit. The impact is amplified by the fact that the system integrates with AI agents like Alexa, Google Home, and Home Assistant. These cameras are the eyes and ears of your AI. When the camera feed is compromised, your AI agent inherits the surveillance exposure. Trust breaks at the vision layer. In a smart contract, if an oracle is corrupted, the redemption price miscalculates. In a home, if the camera is live-streaming to an unknown observer, your digital assistant becomes an unwitting collaborator in your own surveillance.
The financial implications are as massive as the security failure. The global baby monitor market is worth $1.87 billion, with the U.S. accounting for roughly $540 million. Meari itself went public on the Shenzhen ChiNext board in March 2025, seeing its share price double in two days. Retailers continue to sell products that are architecturally incapable of protecting user privacy. This is a different beast than the Zbtlink ENDLESSDOORS vulnerability, which targeted the infrastructure layer, but the structural thesis remains the same: we are buying convenience at the cost of total exposure. Smart contracts don’t lie, people do. But the honest here is that nobody explicitly lied—they just optimized for market speed over security, the same way so many protocols launch with a proxy contract that is upgradeable by a multisig held by people who never check the signatures.
Here is where I have to push back against the “decentralization will fix this” crowd. The usual response to such findings is to recommend local-first alternatives like Home Assistant, or to argue that the IoT cloud should be replaced with a blockchain-based identity registry. That feels elegant because it mirrors the cryptographic trust model. But it’s naive about deployment curves. Home Assistant requires a Raspberry Pi, a YAML config, and a tolerance for breaking changes. The average baby monitor buyer does not have that. And blockchain doesn’t automatically solve the ODM problem either. If Meari deployed a permissioned chain with per-device ACLs encoded as smart contract policies, the vulnerability would shift to the admin key. The whole industry would still have a single point of failure: the company that controls the upgrade mechanism. Entropy increases, but the invariant holds. The invariant is that security is an afterthought when the business model is built on moving cheap hardware with recurring cloud fees.
Moreover, the economic incentives are misaligned. Meari’s initial response to the 70-day coordinated disclosure was to label the affected products as “obsolete.” Meanwhile, the MQTT broker kept streaming. Per the researcher’s disclosure timeline on GitHub, there has been no confirmed GDPR Article 34 direct user notification. This is the same pattern I see in DeFi hacks where a governance emergency fails to pause the contract because the pause function itself is gated behind a timelock. You build a contraption to protect yourself from yourself, and the adversary simply waits for the window you left open. The lack of accountability is exactly why many are turning toward local-first alternatives, where privacy is treated as a competitive moat rather than an afterthought. But that is a niche option, not a systemic solution.
Let me step back and give you a forensics perspective, because I think there’s a deeper lesson here for anyone working in blockchain. During my 2020 audit of a Uniswap V2 fork, I spent 120 hours tracing the swap function’s gas optimization strategies and discovered a subtle arithmetic overflow risk in their custom fee distribution logic. My recommendation was to rewrite the fee mechanism in Rust, which was ignored. The project eventually lost $4 million to a different incident. The point is that developers, both in DeFi and IoT, consistently underestimate the gap between what the code allows and what the user expects. The Meari wildcard subscription is not a cyberattack. It is an architectural permission slip. And it reveals a structural truth about all these systems: the person who deploys the contract holds the admin key, and the user trusts that the admin will act responsibly. But code is law until the reentrancy attack. And here, the reentrancy attack is simply reading the feed you were never supposed to see.
The contrarian angle I want to stress is that this is not a problem blockchain can solve alone. Even if you put the device registry on a public ledger and enforce access via zero-knowledge proofs, you still have the physical layer problem: the camera in the baby’s room is placed there by a parent who trusts the hardware. Verifying the hardware is a supply chain issue, not a consensus issue. The incentive to produce a low-cost camera with adequate security is structurally at odds with the ODM business model, where every dollar spent on security engineering reduces the margin on a $30 device. The market rewards features, not privacy. That’s why I’m skeptical of any blockchain project that claims to decentralize IoT in a meaningful way. You can decentralize the identity layer, but you cannot decentralize the physical deployment of a camera. And you cannot force a parent to read a privacy policy.
In the absence of trust, verify everything twice. That is the only conclusion I can offer. But let me be explicitly forward-looking rather than a comfortable summary. The next critical vulnerability will not be a wildcard subscribe. It will be a genuinely novel attack against the AI-integrated layer—a camera feed that is manipulated or replaced, not just observed. The MQTT stream is the oracle, and when the oracle is compromised, the actions of the AI agent become unpredictable. In DeFi, we quantify oracle risk in price deviations. In the home, the deviation is your toddler’s real-time location. What happens when the wildcard subscription is held by a hostile actor who doesn’t want to watch, but to influence? What happens when they inject a synthetic image into the feed and the AI agent decides to unlock the front door because it thinks your child is crying? We are building the same class of trust problems we already know, just with higher stakes. The invariant holds, but the entropy is still rising.