Partnered With:

From Euler hacker incident, ramble on DeFi security audit and security

Read this article in 6 Minutes

The bounty model, even at 10%, won't beat the hacker pool unless the cost of crime is added to the equation.

Original article by Dforce Mindao

Talk about security audit and security at DeFi through Euler hack incident.

The big DeFi agreements basically go through multiple rounds of audit, we have five audits cost millions of dollars. The regular audit of the big agreements cost millions of dollars every year, but no blue chip DeFi has not been hacked. The reason is very simple, simple math problem. All static audits have limited inputs (security inputs) and outputs (bug discovery).

In addition to regular auditing, Euler also uses Certora for formal validation, which we have used before. Formal validation helps to exhaust the coverage of "known" paths, but not "unknown unknowns." DeFi is an open system with infinite inputs and infinite outputs to the hacker. Suppose the security attack and defense as mining, you use three or five machines hash mining.

The attacking machine is counting hashes all the time, and if it gets it right once, it wins; This win-loss ratio is obvious. Static security audit, because of fixed input and output, can not cover the known unknowns, more can not cover the unknown unknowns. So there is another kind of audit, called exploited-competitive audit, such as Code4rena, where the audit bonus pool is fixed, but the input is elastic for a certain period of time, and anyone can participate, who finds the bug.

According to the severity, bonus points, this way is to let the auditor/white cap to roll, can expand the coverage, but the overall input is still fixed, far from enough. The last is a fully open mode, which is the bounty network. Immunefy, the most famous Immunefy in DeFi, is the platform with the most DeFi white hat masters. I suggest that every DeFi send bounty on it, and the personal test effect is very obvious. Immunefy's grant program is very high.

For example, the highest paid was a Warmhole of $10 million (the same bridge that was hacked for $200 million, and a large bounty). The accident Euler also released a 100w knife bounty, but still did not find this loophole. Bounty mode is also open on input and output, which is similar to hacker attack mode.

But the two incentive models are very different. If the two are taken as a lottery, the same 1000w bonus pool, bounty mode bonus will generally be in 10w-30w knife capped hacker mode is 100% of the bonus away from these two modes, the same input, the same probability of winning, assuming no criminal cost, there is no doubt that the input/output of the hacker pool will win. The bounty model, even at 10%, won't beat the hacker pool unless the cost of crime is added to the equation.

Some people suggest linking the percentage of bounty to TVL, such as 10%, will it motivate more hackers to turn white hat? First of all, no defi protocol can pay a 10% TVL bounty, and second of all, a real hacker is more likely to be willing to hack all the way to the end than to take 10%. DeFi security is further complicated by the fact that in addition to the code level, there are combinable risks (such as the Euler spread across multiple integration protocols).

On the attack surface, DeFi itself increases with integration, and the attack surface grows in four dimensions. Regular static security audit plus long-term bounty cannot cover the ever-expanding attack surface. DeFi security is an infinite game, the only reliable protocol is to reduce external dependence (asset /oracle/ third party protocol), minimize the attack surface, For open systems, the price of security is the price of freedom.