- By Mitchell Amador, founder and CEO of Immunefi
This month, an attacker spent about $4 million to drain roughly $20 million from BonkDAO's treasury. No smart contract failed. The attacker bought enough tokens to pass a governance proposal in a low-turnout vote, and the vote executed exactly as written. The rules themselves were the vulnerability.
This same story repeated in June but from a different angle: the month's largest loss, more than $30 million at Humanity Protocol, came from a private key compromised on a team member's machine, with the contract untouched, per the project's own account.
This is the shape of 2026's worst losses, with crypto losing roughly $972 million so far this year. The number of incidents keeps climbing, and the money increasingly leaves through something other than a contract bug: a stolen signing key, a misconfigured verifier, a treasury anyone can vote their way into. If you look at the sheer number of incidents, you would think the industry is losing ground. But if you look into how much has actually been stolen in total, a narrower, more uncomfortable pattern shows up.
We can be precise about it. Across the 425 hacks we studied from 2021 to 2025, a small share of operational failures carries most of the value lost. In the 2024 to 2025 window, 54.6% of all value lost, across 191 hacks, can be traced to centralized exchange compromises: the keys, custody and signing that sit above the contract.
However, none of this means the code layer is solved. Criticals are everywhere in live code. 93.9% of programs that run five years or more surface a confirmed critical, and roughly one in five confirmed reports is rated critical. The code is never finished either. Every upgrade ships fresh attack surface. What has changed is that continuous, incentivized review now keeps pace with attackers on that code, which is exactly why the same model has to reach further.
This is where the standard playbook runs out. An audit verifies code at a moment in time. It says nothing about who holds signing authority, how a key is stored or what happens when a laptop is compromised. Audits are essential and every serious team should run them, but “we were audited” was never the same as “we are safe.” One protocol was audited 11 times and still lost $128 million.
What actually has hardened contract code is continuous, incentivized pressure. Through live bug bounty programs and a wider stack of monitoring and rapid response, security researchers are paid to find the vulnerability before an attacker does. A roughly $20,000 median bounty routinely prevents a hack that would average around $25 million, making that payout the highest ROI security spend a protocol can make. The model holds because it never stops, and because incentives do not decay when the org chart changes or a signer leaves.
That same discipline now has to cover the keys, the signers and the rules of governance, or we will continue to see catastrophic losses in this area.
So does an audit make a protocol secure? On its own, no. A protocol is secure when its code, its keys, its people, its governance and its monitoring are all treated as a live attack surface, and tested continuously by researchers paid to break them first.
.png?width=1120&upscale=true&name=image%20(24).png)
0 Comments