The crypto industry lost a record $3.4 billion to hacks in 2025, yet the defining story is not about buggy Solidity code. It is about compromised developer laptops, stolen cloud credentials, social engineering campaigns that last months, and multisig wallets with no timelocks.
TL;DR
- Infrastructure and operational failures drove 76% of all crypto hack losses in 2025, while smart contract exploits accounted for just 12%
- North Korean state-sponsored hackers stole $2.02 billion in 2025, roughly 60% of all global crypto theft, using espionage tactics rather than code exploitation
- Audits, bug bounties and on-chain code quality are improving, but the attack surface has expanded far beyond what these tools cover
The numbers say the problem is getting bigger, not narrower
Multiple security firms converge on the same conclusion: 2025 was the costliest year in crypto security history. Chainalysis reported $3.4 billion in stolen funds, up 55% from $2.2 billion in 2024. CertiK documented $3.35 billion across 630 incidents. Fewer attacks occurred than in 2024, but the average payout per incident surged 66.6% to $5.32 million.
The data concentration is extreme. The top three hacks of 2025 accounted for 69% of all service-level losses. The Bybit breach alone stripped $1.46 billion on Feb. 21, 2025, representing roughly 43% of the entire year's theft.
Remove Bybit, and 2025 losses drop to roughly $1.5 billion to $1.9 billion. That is still elevated, but closer to 2024 levels.
The pattern reveals an industry where systemic security has improved for the average protocol, while catastrophic tail risks from infrastructure compromise have grown far worse.
Q1 2025 was the worst quarter in crypto history. Immunefi tracked $1.64 billion across 40 incidents, a 4.7x increase over Q1 2024's $348 million. CeFi accounted for 94% of Q1 losses, driven by just two incidents: Bybit and Phemex ($85 million).
DeFi losses actually fell 69% year-over-year in Q1. On-chain code security genuinely improved even as operational security collapsed.
Early 2026 data shows the trend continuing. CertiK reported $501 million in Q1 2026 losses across 145 events. The Drift Protocol hack on Apr. 1, 2026, drained $285 million in 12 minutes through a six-month social engineering operation. Sophisticated human-targeting attacks remain the primary threat vector entering 2026.
Also Read: Why The U.S. Treasury Is Now Sharing Cyber Threat Data With Crypto Firms
Smart-contract bugs still matter, but they are no longer the whole story
Smart contract vulnerabilities remain the majority of incidents by count, accounting for 54.5% of all exploits. Several significant code-level hacks in 2025 and 2026 proved that traditional on-chain risks persist and continue evolving.
The Cetus Protocol hack ($223 million, May 22, 2025) was a textbook logic error. An integer overflow in a shared math library called "integer-mate" caused an overflow check to silently fail. The attacker minted massive liquidity positions for negligible cost.
Cetus had been audited three times by MoveBit, OtterSec and Zellic. The Zellic audit found zero issues beyond informational items. The vulnerability existed in a third-party dependency rather than Cetus's own code, illustrating how composable ecosystems inherit risks from their entire dependency graph.
Other notable smart contract exploits include:
- The GMX v1 reentrancy ($42 million, Jul. 2025), proving that reentrancy still claims victims through newer cross-contract variants
- The Balancer rounding exploit ($70 million to $128 million, Nov. 2025), which accumulated tiny rounding errors across hundreds of batch swaps in a class of economic attack that standard audits miss entirely
- The Yearn Finance invariant violation ($9 million, Dec. 2025), where a share calculation flaw defeated both static analysis tools and fuzzers
The critical distinction: smart contract exploits tend to produce smaller losses per incident. TRM Labs calculated an average of $6.7 million per code exploit compared to $48.5 million per infrastructure attack. The industry has gotten meaningfully better at writing secure on-chain code. But that progress is overshadowed by the catastrophic scale of operational failures.
Also Read: Stablecoin Volume Could Hit $1.5 Quadrillion By 2035, Chainalysis Report Shows
The human layer: how social engineering became a top-tier crypto exploit
North Korean state-sponsored hackers represent the single largest threat to the crypto industry. Chainalysis attributed $2.02 billion in 2025 crypto theft to DPRK actors, a 51% increase from $1.34 billion in 2024 and approximately 60% of all global crypto theft. The cumulative all-time total reached $6.75 billion through end of 2025.
What makes these operations distinctive is their patience.
The Drift Protocol attack began with conference introductions in fall 2025, progressed through months of relationship building, and included depositing over $1 million of the attackers' own capital to establish credibility. The final drain took 12 minutes.
DPRK tactics have diversified well beyond direct hacking:
- The "Contagious Interview" campaign targets developers through fake job offers on LinkedIn and crypto job boards, deploying trojaned coding challenges that install backdoors
- A fabricated company called "Veltrix Capital" distributed malicious npm packages designed to specifically check for MetaMask browser extensions
- In May 2025, Kraken security identified a North Korean operative applying for an engineering role using the alias "Steven Smith," with voice changes during the interview suggesting real-time coaching
- The IT worker infiltration program generates an estimated $250 million to $600 million annually according to UN estimates, with ZachXBT uncovering a 390-account network generating roughly $1 million monthly
The Coinbase breach (May 2025) demonstrated a different social engineering approach. Bribed overseas customer support contractors exfiltrated personal data from 69,000 users, enabling downstream phishing scams with estimated impact between $180 million and $400 million. The smart contracts were never touched.
Also Read: Cardano Defies Bearish Trend With Record Transactions And Surging Whale Interest
Keys, multisigs and cloud: the hidden centralization inside Web3
The hidden centralization of Web3 is perhaps the industry's most underappreciated systemic risk. Halborn found in its analysis of the top 100 DeFi hacks that only 19% of hacked protocols used multisig wallets and just 2.4% employed cold storage. Off-chain attacks accounted for 80.5% of stolen funds in their dataset.
Trail of Bits published a maturity framework in Jun. 2025 describing four levels of smart contract access control. Level 1 uses a single externally owned account, meaning one compromised private key equals total loss. Level 2 uses a centralized multisig but maintains a single point of control. Levels 3 and 4 add timelocks, role separation and, ultimately, radical immutability.
The Bybit, WazirX and Radiant Capital hacks all exploited Level 2 architectures. The Drift Protocol hack exposed another centralization failure: a 2-of-5 multisig with zero timelocks on any administrative functions.
Cloud infrastructure introduces yet another centralization vector. The Resolv Labs hack ($25 million, Mar. 2026) involved an AWS Key Management Service compromise.
The attacker accessed the cloud environment where a privileged signing key was stored, then used it to mint 80 million unbacked stablecoin tokens.
Resolv had passed 18 independent smart contract audits and maintained a $500,000 Immunefi bug bounty. Neither covered AWS IAM policies.
Many "decentralized" protocols depend entirely on centralized infrastructure providers for their user-facing interfaces. Safe{Wallet}'s frontend was hosted on AWS S3/CloudFront with no Subresource Integrity hashing to detect code modification. That gap enabled the Bybit attack.
Also Read: Bloomberg Strategist Predicts Tether Will Overtake Both Bitcoin And Ethereum By Market Cap
The frontend problem: when users get hacked before the blockchain is touched
A growing category of attacks targets the web frontends of DeFi protocols rather than their on-chain contracts. In every documented case, the smart contracts remained secure and operational. The vulnerability was entirely in the Web2 infrastructure layer connecting users to those contracts.
Curve Finance suffered a DNS hijack on May 12, 2025 when attackers gained access to the domain registrar iwantmyname and modified DNS delegation to redirect traffic to a malicious static decoy site.
During the frontend outage, Curve's smart contracts processed over $400 million in on-chain volume, demonstrating the contracts worked perfectly while the frontend was weaponized.
This was Curve's second DNS attack through the same registrar. Curve subsequently migrated to the curve.finance domain and advocated for industry adoption of ENS.
Aerodrome and Velodrome (Nov. 21, 2025) lost approximately $700,000 when DNS hijacking redirected users to phishing sites. MetaMask and Coinbase Wallet displayed warnings within two minutes of the first malicious transaction, but users who interacted before the warnings lost funds.
Additional DNS attacks hit Arrakis Finance (Jan. 2025), OpenEden (Feb. 2026) and Neutrl (Mar. 2026).
The Neutrl attack was confirmed to have originated from a social engineering attack on the DNS provider itself.
The pattern is consistent: compromise the domain registrar, modify DNS records, redirect users to a phishing clone, harvest wallet approvals, and drain assets. Domain registrars function as centralized single points of failure for nominally decentralized protocols.
Also Read: Cloudflare Follows Google, Sets 2029 Deadline To Go Quantum-Proof
Why audits are not enough anymore
Standard smart contract audits cover code-level vulnerabilities: reentrancy, overflow, access control bugs and known vulnerability patterns.
They typically do not cover frontend and UI security, API and backend infrastructure, admin key management, social engineering vectors, supply chain attacks on dependencies, DNS and domain security, or economic model correctness.
Trail of Bits stated explicitly in Jun. 2025 that private key attacks are an emerging vector that narrowly scoped smart contract audits and contests regularly miss. The firm noted that blockchain-native audit firms rarely flag architectural access control issues as formal findings.
The evidence is extensive:
- Cetus Protocol passed three audits from reputable firms before losing $223 million to a bug in a third-party math library
- Resolv Labs passed 18 independent audits before losing $25 million through an AWS infrastructure compromise
- Bybit's wallet provider Safe{Wallet} was thoroughly audited, but the vulnerability was a developer's compromised laptop
- The Balancer rounding exploit accumulated sub-wei rounding errors across adversarial batch swap sequences, a class of attack that standard per-operation testing cannot detect
Audits remain valuable. Unaudited protocols face roughly a 70% chance of exploit in their first year versus 15% to 20% for audited ones. But the industry's reliance on "audited by X" as a security certification fundamentally mischaracterizes what audits actually validate. They are point-in-time snapshots of code correctness, not comprehensive security assessments.
Also Read: Bitcoin Can Be Made Quantum-Safe Without An Upgrade, But There's A Catch
What safer design looks like after the 2026 hack wave
Vyper, the Pythonic smart contract language created by Vitalik Buterin in 2017, embodies a security-through-simplicity philosophy that contrasts sharply with Solidity's feature richness. Vyper deliberately excludes inheritance, modifiers, operator overloading and inline assembly.
It provides automatic overflow checking, built-in nonreentrant decorators, bounds-validated arrays and strict typing.
Over 7,959 Vyper contracts currently secure more than $2.3 billion in total value locked.
The language faced its own security crisis in Jul. 2023 when a reentrancy guard vulnerability in older compiler versions enabled the Curve Finance exploit. The response was systematic: 12 audits with firms including ChainSecurity and OtterSec, two security experts retained full-time, two bug bounty programs, and a contract monitoring system indexing 30,000 contracts across 23 chains.
Development has remained active through 2025 and 2026. Version 0.4.2 "Lernaean Hydra" (May 2025) notably banned calling nonreentrant functions from within nonreentrant functions, eliminating an entire class of potential vulnerabilities.
Major adopters include Curve Finance, Yearn Finance V3 and Velodrome/Aerodrome.
Vyper's design philosophy, that the features you remove matter more than the features you add, aligns with the emerging security consensus. When the dominant attack vectors are human and operational rather than code-level, a language that produces more readable and auditable code offers genuine structural advantages.
Also Read: Only 10% Of New CEX Tokens Survive Their First Year, CoinGecko Data Reveals
Security culture vs security theatre in Web3
Bug bounties have emerged as one of the more cost-effective defenses. Immunefi has paid over $112 million in total bounties across more than 3,000 validated bug reports. Critical severity reports account for 87.8% of all payouts. The platform claims to have protected over $190 billion in user funds.
The economics make a compelling case. Total bounties ever paid ($112 million) represent roughly 3.3% of 2025's hack losses alone. Even a single prevented exploit would generate enormous ROI. Active bounty programs now reach meaningful scale, with Usual offering $16 million maximum on Sherlock and Uniswap v4 offering $15.5 million on Immunefi.
Competitive audit platforms have evolved alongside traditional bounties. Code4rena runs contests with 16,600 registered researchers and roughly 100 participants per audit.
Sherlock operates a full lifecycle model combining audit contests, bug bounties and insurance coverage, having secured over $100 billion in total value locked.
However, bug bounties share a fundamental limitation with audits. Immunefi's data shows that 77.5% of payouts go to smart contract bug discoveries. The most damaging attack vectors of 2025, including supply chain compromises, social engineering and infrastructure breaches, fall largely outside the scope of what bounty researchers can test for.
The industry needs equivalent incentive structures for operational security assessment. Code review alone no longer matches where the actual losses originate.
Also Read: Gemini Survey Reveals 51% Of Gen Z Adults Worldwide Own Crypto
What users, builders and investors should do differently in 2026
The 2025 and 2026 data make clear that security must extend far beyond smart contract audits to encompass the full operational stack.
For protocol builders, the minimum viable security posture now includes:
- Multisig wallets with timelocks on all administrative functions
- Role-based access control with principle of least privilege
- Hardware wallet signing for all privileged operations
- Continuous monitoring for permission changes, upgrades and high-value transfers
- Subresource Integrity hashing for all frontend code, plus DNSSEC and consideration of ENS-hosted alternatives to centralized domain registrars
Supply chain security requires pinning dependencies, limiting package sprawl, locking CI/CD pipelines with short-lived credentials and verifying release artifacts. Incident response plans should be tested with drills, not written and shelved.
For users, the practical defenses are straightforward. Hardware wallets remain essential for any significant holdings. Transaction simulation tools like Pocket Universe claim 180,000 users and over $1 billion in protected funds.
Regular revocation of unlimited token approvals, bookmarking trusted URLs rather than following links, and maintaining separate wallets all reduce the blast radius of a single compromised signature.
The Bybit hack's blind signing lesson applies to individuals too. Always verify transaction details on the signing device itself, not just the interface requesting the signature.
For investors evaluating protocols, the "audited by X" label is necessary but radically insufficient. Meaningful security indicators include multiple audits from diverse firms, active bug bounty programs with significant rewards, transparent multisig configurations with geographic distribution, timelocks on upgrade functions visible on-chain, and demonstrated incident response capability.
The absence of these indicators should function as explicit red flags regardless of audit history.
Also Read: From Joke To ETF? PEPE's Wall Street Moment Raises Big Questions
Conclusion
The crypto security landscape of 2025 and 2026 reveals a paradox. The technology is getting more secure while the industry is losing more money than ever.
Smart contract exploit losses have declined as on-chain code quality improves through better tooling, more audits, competitive review platforms and security-conscious language design like Vyper. But this progress has been overwhelmed by a rapid escalation in infrastructure and operational attacks.
Security models built around code review alone now address roughly 12% of actual loss exposure. The remaining 88% lives in developer laptops, AWS credentials, domain registrars, multisig signer devices, employee hiring processes and frontend deployment pipelines. These are Web2 problems requiring Web2 defenses applied to Web3 organizations that often lack the institutional security culture to implement them.
The protocols that survive the next wave of state-sponsored attacks will be those that secure not just their code, but their people, their infrastructure and their trust assumptions as one connected system. Anything less is security theatre dressed in decentralization branding.
Read Next: The U.S. Is Redefining Stablecoins — Here's What The New Rules Do