Crypto Security Report 2025–2026: $3.4 billion stolen, state actors, and what changed

1. The numbers: 2025 by the data

The headline figure — $3.41 billion in total losses — tells only part of the story. The structure of those losses changed fundamentally in 2025.

While the total incident count dropped from 410 in 2024 to approximately 200 in 2025, the average loss per incident more than doubled. Q1 2025 was the worst quarter in crypto history at $1.64 billion, driven almost entirely by the Bybit hack. Q2 brought $801 million in losses, while Q3 settled to $509 million as the industry adapted. September 2025 set a record of 16 separate "million-dollar hacks" in a single month, demonstrating that even as mega-hacks drew attention, the long tail of mid-sized attacks continued unabated.

DeFi losses fell to $649 million, a 37% decline that reflects genuine improvement in smart contract auditing practices and bug bounty programs. But centralized exchange losses surged to $1.8 billion — a massive increase driven by the Bybit and Coinbase incidents. Wallet compromises were the most efficient attack category: just 34 incidents accounted for $1.71 billion, or 69% of all value lost.

On a more positive note, fund recovery improved. A total of $387 million (13.2% of losses) was recovered through on-chain freezes, law enforcement coordination, and protocol governance actions — up from previous years in both absolute and percentage terms.

2. The Bybit mega-hack: $1.5 billion

On February 21, 2025, the Dubai-based exchange Bybit suffered the largest single hack in cryptocurrency history. A total of 401,347 ETH was drained from Safe-based multisig wallets spanning Ethereum and Arbitrum. The loss — approximately $1.5 billion at the time — represented roughly 50% of all crypto theft in 2025.

The root cause was not a smart contract vulnerability. The attackers compromised internal signing keys through a months-long campaign combining social engineering and malware deployment against key employees. By gaining control of enough signing keys, they were able to authorize massive transfers without triggering the exchange's anomaly detection systems.

Post-theft, the funds were rapidly fragmented across newly generated addresses, moved through cross-chain bridges, and processed through mixing services. The attack was attributed to North Korea's Lazarus Group based on on-chain behavioral patterns, laundering infrastructure overlap, and intelligence community assessments.

The Bybit incident carries a critical lesson for the industry: multisig security is only as strong as the operational security of every signer. When the signing environment — the combination of people, devices, and processes — is compromised, the cryptographic protections become irrelevant. For a deeper look at how multisig works and its limitations, see our dedicated guide.

3. Coinbase: $400 million social engineering attack

In May 2025, Coinbase disclosed an attack that stood apart from every other incident in this report: it involved no technical exploit whatsoever. The attackers bribed and psychologically manipulated internal customer support employees over an extended period, gradually gaining access to administrative support tools.

With that access, the attackers could view customer account screenshots, extract credentials, and bypass authentication mechanisms. The breach exposed customer data and led to downstream theft. Coinbase allocated up to $400 million in customer compensation.

The Coinbase attack underscores a reality that the crypto industry has been slow to acknowledge: human-operated processes are often the weakest link in otherwise technically impenetrable defenses. No amount of smart contract auditing or cryptographic security can compensate for compromised internal personnel. For users, this reinforces the value of self-custody for significant holdings — your assets on an exchange are only as secure as the exchange's least security-conscious employee.

4. Major DeFi exploits of 2025

Despite the overall decline in DeFi losses, several exploits in 2025 demonstrated that protocol-level attacks remain a serious threat — particularly when they target subtle mathematical errors or novel chain architectures.

Cetus Protocol — $223 million

In May 2025, Cetus Protocol, the largest decentralized exchange on the Sui network, was hit by an exploit rooted in an integer overflow error in its liquidity calculations. The attacker created fake tokens mimicking legitimate assets and exploited vulnerabilities in the protocol's asset handling logic to manipulate on-chain prices. Liquidity providers were left holding worthless positions as the attacker drained the pools. The incident highlighted that newer blockchain ecosystems like Sui face the same fundamental security challenges as Ethereum and others — the novelty of the chain does not confer immunity.

Balancer V2 — $128 million

In November 2025, Balancer V2 suffered an exploit across multiple chains due to a rounding error in its composable stable pool mathematics. The attacker crafted sequences of swaps that exploited tiny discrepancies between the protocol's internal accounting and actual pool balances. Each individual rounding error was negligible, but by executing thousands of precisely calibrated transactions, the attacker extracted $128 million. This type of attack is extremely difficult to detect through standard auditing because the underlying math is correct in isolation — the vulnerability only emerges in the interaction between rounding and repeated operations.

GMX V1 — $42 million

In July 2025, GMX V1 on Arbitrum was exploited through a reentrancy vulnerability. The attacker minted excess GLP tokens and manipulated the protocol's AUM calculations by executing short positions directly from the vault contract, bypassing the ShortsTracker entirely. The exploit demonstrated that even battle-tested protocols on Layer 2 networks can harbor latent vulnerabilities, particularly in legacy code that predates more recent security patterns.

For a complete history of major crypto exploits, including the bridge attacks of 2022 and the exchange failures of 2023–2024, see our comprehensive timeline of crypto hacks.

5. State-sponsored attacks: the geopolitics of crypto theft

Cryptocurrency theft is no longer a story about anonymous hackers and opportunistic criminals. In 2025, the most significant threat actors were nation-states using crypto theft and sanctions evasion as instruments of foreign policy and military funding.

North Korea — $2 billion in 2025

North Korea's crypto theft operations reached $2 billion in 2025, a 51% increase over 2024. The Lazarus Group, the primary state-sponsored hacking organization, has refined its approach from direct exchange hacks to a more insidious model: thousands of IT workers operating globally under fake identities, infiltrating crypto, Web3, and AI companies to gain privileged access from the inside.

These operatives apply for engineering and DevOps roles, pass technical interviews, and work at companies for months or years before exploiting their access. Once inside, they can exfiltrate keys, install backdoors, or directly authorize malicious transactions. The stolen funds follow a predictable 45-day laundering cycle through Chinese-language services, cross-chain bridges, and specialized mixers. Notably, Lazarus Group avoids DeFi lending protocols and prefers low-KYC exchanges for cash-out.

For context on how North Korea launders stolen crypto through bridge infrastructure, see What is a crypto bridge?

Russia — the A7A5 stablecoin

Russia's crypto activity in 2025 centered on the A7A5 stablecoin, a ruble-linked token that processed $72 billion in total volume (an estimated 34% of which was wash trading). A7A5 functions as a bridge for Russian businesses to access global supply chains while circumventing Western sanctions. Investigations linked A7A5 transactions to the acquisition of missile components through intermediaries in China and Kyrgyzstan.

Sanctions-linked crypto transactions surged 694% in 2025, reaching $154 billion in total illicit volume across all Russian-linked entities. For more on the risks that stablecoins carry beyond price stability, including sanctions exposure and counterparty risk, see our stablecoin risks guide.

Iran — IRGC dominance

Iran's Islamic Revolutionary Guard Corps (IRGC) dominated Iranian crypto activity in 2025. In Q4 alone, IRGC-linked addresses received more than 50% of all value flowing to Iranian entities on-chain. The IRGC moved over $3 billion through crypto channels to fund regional militia networks and acquire dual-use equipment.

In February 2026, US-Israeli military strikes on Iranian infrastructure triggered real-time on-chain asset movements from IRGC-linked wallets — a striking example of blockchain data serving as a barometer of kinetic military conflict. On-chain intelligence has become a tool of geopolitical analysis in its own right.

6. New attack vectors: EIP-7702 and account abstraction risks

Ethereum's Pectra upgrade in May 2025 introduced EIP-7702, a technical change with significant security implications that are still being understood. EIP-7702 allows externally owned accounts (EOAs) — standard user wallets — to temporarily act as smart contracts by delegating execution to a contract address.

This creates several novel attack surfaces:

• Broken reentrancy protections. The long-standing guard require(tx.origin == msg.sender), used by many contracts to prevent reentrancy, is no longer reliable. An EOA that has delegated to a contract can now be both the transaction origin and a contract executor simultaneously.
• Whitelist bypasses. Contracts that whitelist specific EOA addresses assume those addresses cannot execute arbitrary code. With EIP-7702, a whitelisted EOA can become a programmable gateway, executing any logic its delegate contract contains.
• Address detection failures. The extcodesize check, commonly used to determine whether an address is a contract, now returns unreliable results for delegated EOAs. Security logic that depends on distinguishing between user wallets and contracts can be defeated.

These risks are not theoretical. Security researchers demonstrated proof-of-concept exploits within weeks of the upgrade. Any protocol that has not audited its contracts against EIP-7702's behavioral changes is operating with assumptions that may no longer hold. This is a novel attack surface that simply did not exist before May 2025.

7. Early 2026: signs of maturity — and new threats

February 2026 brought a striking statistic: a 98.2% drop in hack losses compared to February 2025 ($26.5 million vs $1.5 billion+). While the comparison is skewed by the Bybit outlier, the broader trend in early 2026 suggests improved industry preparedness — at least against the attack vectors that dominated 2025.

But Q1 2026 also brought new incidents that demonstrated how the threat landscape continues to evolve:

The $282 million Trezor user theft demands particular attention. An individual holder was convinced by scammers posing as support agents to reveal their seed phrase. This is now the most expensive social engineering attack against an individual in crypto history. Hardware wallets provide strong protection against malware and remote key extraction — but they offer zero protection against a user voluntarily surrendering their recovery phrase. The device cannot distinguish between a legitimate recovery and a social engineering attack.

The TrueBit and YieldBlox incidents also illustrate an ongoing problem: legacy contracts with known vulnerability classes (integer overflows, oracle manipulation) remain deployed and holding funds. New chains like Stellar are not immune to attack patterns that Ethereum-based protocols have learned to guard against. For more on how people lose crypto through both technical and human failures, see our guide.

8. AI-powered fraud: identity theft 2.0

2025 saw an 89% increase in AI-enabled attacks across the crypto ecosystem. The convergence of large language models, face-swap deepfake technology, and voice cloning has fundamentally changed the economics of social engineering.

AI-powered scammers now deploy face-swap deepfakes for real-time video calls and use LLM-simulated support agents that can maintain coherent, context-aware conversations for hours. These tools extract 4.5x more money per successful attack than traditional human scammers. The volume is equally concerning: AI-enabled operators execute up to 35 fraudulent transfers per day, compared to 3.8 for human-only scammers.

An emerging threat that security researchers flagged in late 2025 is "AI agent poisoning" — the injection of malicious data into automated trading agents or DeFi bots to redirect funds. As more users delegate transaction execution to AI agents, the attack surface expands: a poisoned price feed or manipulated training dataset can cause an agent to execute transactions that benefit the attacker. This risk is compounded by the fact that most AI agents operate with broad token approvals, giving them permission to move assets without per-transaction user confirmation.

For practical steps to defend against these threats, see Staying safe in crypto.

9. The NFT infrastructure collapse

While not a hacking story in the traditional sense, the NFT infrastructure collapse of 2025–2026 represents a different kind of security failure: the loss of asset integrity through platform abandonment.

NFT supply grew 25% in 2025, but sales volume dropped 37%, with the average sale price falling below $100 for most market segments. The economic pressure proved fatal for several major platforms: Nifty Gateway closed in February 2026, followed by exits from Kraken NFT and Bybit NFT.

The deeper problem is structural. Analysis shows that 27% of major NFT collections depend on centralized servers for their metadata — the images, descriptions, and attributes that define what an NFT actually represents. When these platforms shut down, the metadata links break. Thousands of NFTs now point to dead URLs. The tokens still exist on-chain, but they reference nothing. Millions of dollars in assets have been reduced to empty digital pointers.

This is a form of risk that most NFT buyers never considered: the token's immutability on the blockchain does not extend to the off-chain content it references. True permanence requires on-chain or decentralized storage (IPFS with pinning, Arweave), which most collections never implemented.

10. Regulatory response: the era of mandatory transparency

The scale of 2025's losses accelerated regulatory timelines worldwide. The most significant development is the EU's Markets in Crypto-Assets (MiCA) regulation, which enters full application on June 1, 2026.

MiCA requires formal licensing for all Crypto-Asset Service Providers (CASPs), audited capital reserves, and strict asset segregation rules. The regulation mandates that customer funds be held separately from corporate assets — a direct response to the FTX model where customer deposits were comingled with trading operations.

Beyond MiCA, several security-focused regulatory trends gained momentum in 2025:

• Cryptographic proof of reserves — Merkle tree-based attestations and on-chain reserve proofs are replacing traditional sworn statements. Exchanges that can demonstrate real-time solvency gain a competitive advantage.
• SOC 2 Type II certification — becoming the baseline security standard for institutional crypto infrastructure, covering access controls, change management, and incident response.
• Real-time monitoring and protocol pausing — the Venus Protocol model (September 2025) showed what's possible: Hexagate monitoring detected an incoming attack 18 hours before execution, the protocol was paused through governance action, and funds were recovered. This "defense-in-depth" approach is becoming a template.
• Travel Rule enforcement — the EU now requires identity information for crypto transfers exceeding €1,000, aligning crypto with traditional banking wire transfer requirements.
• Compliance-by-design — OFAC sanctions monitoring is increasingly built directly into smart contracts and user interfaces, preventing sanctioned addresses from interacting with protocols at the contract level.

11. What this means for you

The data from 2025 tells a clear story: the attack surface has shifted decisively from code to people and operations. For individual users and portfolio managers, the practical implications are significant.

Your biggest risks are not where you think they are. Forgotten token approvals, compromised signing environments, and social engineering are now more dangerous than smart contract bugs. If you have ever granted unlimited token approvals to a protocol you no longer use, those approvals remain active and exploitable. Review and revoke them.

Hardware wallets protect against malware and remote key extraction, but they cannot protect you from revealing your seed phrase to a convincing impersonator. The $282 million Trezor user theft is the most expensive illustration of this distinction. No hardware device can override a user's decision to hand over their recovery phrase.

Self-custody combined with real decentralization remains the strongest defense model — if you manage the human risk. The protocols that survived 2025 intact were those with genuine decentralization: distributed key management, time-locked governance, and no single point of human failure. For guidance on evaluating these properties, see Understanding risk in DeFi.

Tools like CleanSky exist to help you see your full exposure: all positions, all token approvals, all protocol risks across every chain — in a single view. When the attack surface is this broad, visibility is the first line of defense.

The institutional outlook is also relevant for context. By end of 2026, 76% of institutional investors are expected to increase their digital asset exposure. The capital flowing into crypto is growing, which means both the value at risk and the sophistication of attackers will continue to increase. The security landscape of 2025 is the new baseline, not the peak.