In July 2026, the BonkDAO, which governs the meme coin on Solana, operated flawlessly. However, approximately $20 million was siphoned from its treasury. No hacking occurred in the traditional sense; rather, the voting process was manipulated legitimately.

This article from ForkLog explores how governance attacks are becoming increasingly common, with human factors being the primary target for hackers. Additionally, AI analysis of legacy smart contracts poses a dual threat to cybersecurity and serves as a tool for malicious actors.

Decentralization Under Attack

On July 6, 2026, BonkDAO faced a governance takeover, resulting in the theft of around 4.43 trillion BONK tokens from its treasury. The attacker exploited inherent weaknesses in the DAO's architecture and the community's inactivity, executing a transaction that was legal from a coding perspective.

The attack was disguised as a routine initiative, Bonk Improvement Proposal #76, which aimed to reward voting participants. The hacker incorporated a malicious function that redirected funds from the treasury to their own address.

To secure enough votes, the attacker spent approximately $4.4 million on centralized exchanges, acquiring just one percent of the total BONK supply needed to reach a quorum. Given the extremely low voter turnout, this was sufficient to control nearly 100% of the votes. Six days later, the proposal was executed automatically.

This incident exposed critical vulnerabilities in the DAO's governance system. There was no mechanism for delaying decision implementation, which would allow developers to cancel abnormal transactions. Furthermore, the quorum required to manage such substantial sums was too low, and the team lacked emergency veto rights.

The attacker scrutinized the organization's charter as closely as any diligent participant and identified a weak point.

This was not an isolated incident. A year earlier, a similar attempt was made against the Compound protocol's DAO. In July 2024, hackers tried to withdraw 499,000 COMP tokens (worth about $24.1 million at the time) from the protocol's treasury through a malicious improvement proposal. With typical voter turnout at around 4-5% of the total token supply, this package was enough to gain control over the voting process. Unlike the Bonk case, this attack was detected and repelled, albeit after the proposal had passed, through negotiations.

Low voter turnout and token concentration in a few hands are not merely issues of code exploitation; they highlight flaws in the decentralization mechanisms within Web3.

In early 2026, Ethereum co-founder Vitalik Buterin urged the crypto community to develop more advanced and effective DAO structures.

He criticized the prevailing token-based governance model as unproductive and weak, stating that while it functions, it is inefficient, vulnerable to capture, and "fails to mitigate the flaws of human politics." This is a key reason for the community's skepticism toward such organizations.

Are Audits Enough?

According to SlowMist, the number of incidents in Web3 fell nearly by half in 2025—down to about 200 from 410 the previous year. However, total losses increased from $2 billion in 2024 to $2.9 billion in 2025.

Experts at Chainalysis estimated the losses from hacks to be even higher, totaling $3.4 billion. They noted that three incidents accounted for 69% of all losses—marking the first time in history that the gap between the largest attack and the average was a staggering 1,000-fold.

While DAO governance seeks out vulnerabilities without a single line of exploit, issues within the code are different: technical bugs are well-known but were somehow not flagged in advance.

According to analysts from Kerberus, 90% of the smart contracts hacked in 2025 had passed audits successfully. Most successful attacks did not require the discovery of complex zero-day vulnerabilities; instead, hackers employed techniques that have been known in the industry for years:

  • Reentrancy errors—where a function is called again before the previous call has completed;
  • Access control errors—where a critical security function is accessible to more individuals than intended;
  • Oracle manipulation—distorting asset price data relied upon by smart contracts;
  • Logical errors within the protocol's business system that scanners fail to detect.

Cross-chain protocols are particularly in need of thorough code audits.

According to Phemex, as of April 2026, bridges have collectively lost over $2.8 billion since 2022—about 40% of all funds stolen in Web3. This is due to the complexity of cross-chain messaging, which creates a wide attack surface that is challenging for specialists to assess completely; numerous independent components must operate in parallel, and a mistake in just one can lead to a breach.

Major DeFi hacks from January 1 to April 19, 2026. Source: Phemex.

The year 2026 showcased several illustrative examples with varying attack vectors:

  1. On January 8, the Ethereum verification protocol Truebit lost $26.4 million due to integer overflow in its code. The attack targeted an outdated smart contract for buying/releasing TRU tokens deployed five years prior—a classic logical error that could have been detected had the audit been redone post-update.
  2. On February 21, hackers gained control of the validator's private key for the IoTeX bridge, updated the contract to a malicious version bypassing signature checks, and withdrew $4.4 million while minting over 111 million CIOTX and CCS tokens.
  3. On April 18, the largest DeFi exploit of the year occurred when the liquid restaking protocol Kelp was targeted: around 116,500 rsETH worth $292 million was stolen from the LayerZero bridge. This incident was made possible by a configuration with a single validator for cross-chain messaging and an attack on LayerZero's RPC infrastructure.

Badges from cybersecurity organizations like CertiK and Code4rena have not been able to safeguard assets in this still young industry.

New attack vectors require a rethinking of defense mechanisms, which cybersecurity specialists are already implementing, despite challenges posed by advancements in artificial intelligence.

Prompt Over Auditor

While experts examined recurring patterns of vulnerabilities, they faced a new adversary for whom such analysis is a routine task performed at speeds beyond the capabilities of any team of human experts.

On December 1, 2025, Anthropic announced the use of AI models to identify gaps in smart contracts, uncovering new exploits totaling $4.6 million.

To conduct their evaluation, programmers assembled a benchmark called SCONE-bench, consisting of 405 smart contracts that were actually hacked between 2020 and 2025. During simulations, the neural networks successfully attacked 207 of these contracts, extracting a total of $550.1 million. A significant portion of these vulnerabilities may have been accessible to LLMs from training data.

To mitigate the risk of "memory replay," Anthropic separately tested 34 smart contracts that were hacked after March 2025.

A logarithmic graph of cumulative returns from successful exploitation of smart contract vulnerabilities by Anthropic's models in 2025. Source: Anthropic.

This phenomenon is already occurring. In June, Chainalysis reported abnormal activity in networks—multiple nearly identical attempts to hack different contracts simultaneously, indicating automated vulnerability probing rather than manual work by individual hackers who could not physically attack dozens of targets at once.

The entry threshold for hacking is decreasing much faster than the quality of defenses is improving. Just a few years ago, finding an exploitable vulnerability in someone else's contract required an experienced Solidity specialist capable of manually reading the code and understanding its logic. Based on the SCONE-bench results, for a significant portion of standard attacks, this task may now be reduced to simply formulating the right query.

Keys Matter More Than Contracts

No audit of a smart contract could have prevented the largest thefts of the last two years.

According to a Chainalysis report from 2024, private key theft accounted for about 40% of stolen funds—more than any other individual attack vector, including smart contract vulnerabilities.

Since 2020, attacks on crypto infrastructure have increasingly targeted not just blockchains but the services and infrastructure surrounding them—exchanges, wallets, and interfaces through which individuals manage their assets.

A notable example is the Bybit hack. On February 21, 2025, the exchange lost $1.46 billion in ETH, marking the largest single theft in crypto history. Hackers intervened in the process of a routine transfer from a cold wallet to a hot one: in the Safe interface, signers saw altered data and approved what appeared to be a legitimate transaction, which effectively transferred control of the wallet to the cybercriminals.

In June 2026, a similar situation befell Humanity Protocol: hackers accessed keys through a backup stored on a developer's computer infected with malware. The losses amounted to approximately $31 million.

What Works?

When the breach occurs not in the code but in the decision-making process, mere audits of smart contracts are no longer sufficient. Hence, the industry is evolving its tools and approaches to storing and verifying operations:

  • MPC wallets—cryptocurrency storage systems utilizing multi-party computation technology. Instead of one complete secret key, they create several independent parts that are stored separately. This eliminates a single point of failure and simplifies access recovery;
  • Account abstraction with a social recovery access mechanism—loss or theft of a key no longer means irrevocable loss of funds;
  • Passkey authentication—a method for account access without a password, utilizing cryptographic key pairs and confirming identity through biometrics (fingerprint or facial scan) or device PIN. This counters phishing;
  • Utilization of hardware and cold crypto wallets in situations previously reliant on software solutions on employees' workstations.

Each of these mechanisms would have been beneficial in the Bybit and Humanity Protocol cases—both incidents stemmed from software and workplace device compromises.

Protection is gradually shifting from one-time checks to dynamic, real-time threat monitoring. Projects are implementing automated systems such as Forta Network and Hypernative, which track abnormal activity in the mempool before a malicious transaction is included in a block.

This proactive response model has already demonstrated its effectiveness. On September 2, 2025, when hackers attempted to attack the lending protocol Venus, Chainalysis's Hexagate monitoring system detected the deployment of a suspicious smart contract and abnormal activity 18 hours prior to the main attack phase. Twenty minutes before the final transaction, which would have drained the pools, the protocol automatically paused. The funds remained secure, and the hackers wasted resources on gas fees.

The architecture of Web3 was originally designed with a radical goal—to entirely eliminate trust in human beings and transfer control to immutable mathematics. However, the statistics of breaches in recent years illustrate the opposite effect. Perfectly written and double-checked code proves utterly powerless if a developer clicks on a phishing link, and a decentralized community remains unaware during a DAO takeover. Perhaps the primary systemic vulnerability in blockchain economics lies not in the algorithms but in the fact that people are still at both ends of the wire?