Budget for the boring controls first
Security theater loves a clever bug. The losses cited in Monad Foundation’s own guide point somewhere duller, and they come with a limit attached.
ChainLight’s Lumos tracker, as cited in the Crypto OpSec Bible and reported by crypto.news, recorded $3.31 billion in exploited value from control hijacking, compared with $983 million from smart-contract vulnerabilities. The count starts in January 2023 and stops at a cutoff the coverage does not name. crypto.news presents it as a historical snapshot of incidents that tracker recorded. The outlet does not present it as an industry total.
Control hijacking, in the guide’s categories, means social engineering, private-key leakage, unauthorized access, supply-chain attacks, and other conventional security failures. Code flaws still matter. They were not where most of the recorded money went.
For a leadership team, the claim that travels is this. In that snapshot, the expensive failures are access, credentials, and dependencies. An AI testing agent does not repair that. Run it with broad permissions on a machine that can still sign, and the agent becomes another privileged account.
A renewed push, not a new product
Monad co-founder Keone Hon was not announcing a launch. crypto.news dates an embedded post from @keoneHD to October 30, 2026. He pointed teams back at resources already on the shelf, and he framed the urgency around attackers with better tools.
In this age of AI-empowered attackers, the only way forward for our industry is to continue hardening. Security should be every team’s top focus.
That is his forecast. The reporting does not show that more capable models have raised attacker success rates. It shows a founder telling his industry to harden accounts, devices, and signing paths, and offering an open-source agent for contract testing.
He was direct about the tool: “Smart contract developers, please run Ultrafuzz on your contracts!! It’s open source and available to anyone.” He also sent readers to Omar’s Crypto OpSec Bible, guidance on administrator accounts, and research on finding vulnerabilities with AI, including formal verification and threat hunting with large language models.
Omar’s guide starts from the same compromises showing up again: unfamiliar code, phishing, exposed credentials, and sloppy device management. His argument is that crypto organizations have paid too little attention to operational security relative to code flaws. The Lumos snapshot is the dataset that argument leans on.
The practical recommendations will sound familiar to anyone who has run a company, which is why they get skipped. Separate personal activity from work devices. Keep a clean computer or browser for high-value activity. Read the transaction before you sign it. For critical accounts, use hardware security keys with a securely stored backup, unique passwords, and authentication codes kept apart from those passwords. For organizations, run endpoint protection and name a clear contact for reporting incidents.
Hon also highlighted a narrower rule for people who administer smart contracts: do not rely on ordinary wallet keys, and use a dedicated laptop when signing administrative functions. An admin key living in the same browser profile as everyday browsing is an open drawer.
Keep the agent off the machine that can sign
crypto.news places Ultrafuzz’s open-source announcement on September 10, weeks before those October posts. The public repository is monad-developers/ultrafuzz. It supports Solidity and Vyper, two languages used to write smart contracts. Monad says more than 20 runs on ecosystem projects helped shape the release. The coverage presents that as development history. It does not present a published measure of accuracy.
Antonio Viggiano, whom Hon credits, wrote a technical walkthrough and built the workflow while helping a team prepare for an external security review. That project had no fuzz tests. Fuzzing means throwing large volumes of generated inputs at software to see what breaks. Viggiano describes Ultrafuzz as a further step into AI automation: specialized agents identify conditions a protocol should preserve, test whether actions or sequences break those conditions, then combine, check, and organize the findings into a report.
Anyone buying AI agents for business work will recognize the pattern. Define what must stay true, probe it, and hand a person a report instead of a pile of logs. Security review is one of the few places that pattern can pay for itself quickly.
Monad’s warning belongs next to the pitch. The agents run with unrestricted permissions. Users are told to run Ultrafuzz on disposable, isolated virtual machines, and not on developer workstations or any machine holding valuable credentials. That is the AI risk the reporting actually supports: an agent that can do anything, on a computer that can still move money.
Three breaches, none of them an AI case file
crypto.news sets three incidents beside the guide as examples of account, credential, and software-dependency risk. Neither that outlet nor The Defiant says these cases were driven by AI. They show how control gets taken without a brilliant contract bug.
On October 26, crypto.news reported that wallet maker Foundation suffered a Cloudflare API key compromise, so fraudulent emails appeared to come from official domains. Foundation revoked the credentials and said Passport hardware and stored funds remained secure. The Defiant added that recipients saw personal and purchase details in the messages. A cloud API key is an admin console with better manners: once it can send mail as you, the phish does not need to look homemade.
In July, as crypto.news reported, security firm Socket said a compromised Injective developer package intercepted wallet-generation functions and sent private keys and recovery phrases through fake telemetry. Injective CEO Eric Chen said the affected releases had been deprecated and the issue fixed, and no loss figure is given. Your dependency tree is part of production whether or not you staffed it that way.
Another July case, which crypto.news reported from a Kaspersky exposure, was OkoBot: wallet-stealing malware in GitHub repositories disguised as legitimate software, pushed with fake error messages, repair instructions, fraudulent wallet-recovery screens, keyboard monitoring, and clipboard surveillance. The repo looked helpful. The install was the breach.
A few things stay open. What specifically prompted Hon’s posts beyond “AI-empowered attackers.” How Lumos defined and verified those dollar figures. Whether Ultrafuzz adoption rose after October 30. And what the formal-verification and large-language-model threat-hunting notes actually recommend.
Questions worth asking before you adopt the pitch
-
What do the $3.31 billion and $983 million figures prove?
Less than a clean industry ranking. ChainLight’s Lumos tracker, as cited by Monad’s guide and reported by crypto.news, recorded $3.31 billion in exploited value from control hijacking and $983 million from smart-contract vulnerabilities, from January 2023 to an unnamed cutoff. It is a historical snapshot of incidents that tracker logged. It is not a current or complete industry total, and the methodology is not independently verified here.
-
Why does Ultrafuzz need an isolated machine?
Its agents run with unrestricted permissions. Monad tells users to run it on disposable, isolated virtual machines and to keep it off developer workstations and any machine holding valuable credentials.
-
Which few controls come first?
The guide’s concrete list, as reported by crypto.news: separate personal activity from work devices, keep a clean computer or browser for high-value activity, read the transaction before signing, use hardware security keys with a stored backup, keep unique passwords apart from authentication codes, run endpoint protection, and name a contact for incidents. For contract administrators, do not rely on ordinary wallet keys, and sign administrative actions from a dedicated laptop.
Budget for the boring controls first. Then let AI agents hunt bugs inside a fence, not from the laptop that can still sign.