Key Takeaways
- H1 2026 recorded the highest-ever crypto hack incident count (207+), yet total losses fell below $1 billion for the first time since 2022 — a misleading signal, since infrastructure compromises representing just 15% of incidents drove 76% of total losses.
- AI adoption in crypto crime nearly doubled in a single year. Deepfake scam losses in 2026 have already exceeded the full-year 2025 total, and the first fully autonomous AI ransomware operation is now documented.
- The Term Finance governance exploit showed that code-level audits and governance timelocks alone don’t prevent attacks targeting the operational layer around the code.
- Defending against this shift requires two things working together: custody architecture that eliminates single points of failure, and real-time transaction intelligence that catches unauthorized movement before settlement.
H1 2026 Crypto Hacks Overview: The Shift in Loss Concentration
The first half of 2026 produced a misleading headline figure.
Crypto hack incidents hit a record, with over 207 separate exploits verified by Immunefi and higher counts reported by Blockaid and CertiK using different methodologies. Yet total losses came in below $1 billion for the first time since 2022, according to Immunefi’s June ecosystem update. DeFi-specific exploit damage has fallen 74% from its 2022 peak.
On the surface, this suggests improving industry defense: more attacks intercepted, less capital lost, protocols maturing. The composition of losses tells a different story.
According to TRM Labs’ H1 2026 analysis, infrastructure and operational compromises represented only about 15% of total incidents but accounted for roughly 76% of total losses. The decline in per-incident damage comes almost entirely from smaller smart contract exploits being caught earlier and draining less. Large-scale infrastructure breaches, the ones producing eight- and nine-figure losses, have not decreased in severity. They have become more concentrated.
That concentration is the critical development. A reduction in catastrophic events does not indicate a reduction in risk. It indicates that risk has consolidated into a smaller number of higher-impact targets, where a single compromised signing workflow or governance mechanism can drain a protocol faster than a hundred smart contract vulnerabilities combined.
The Term Finance Exploit: A Governance Problem, Not a Code Problem
On August 23, 2026, Ethereum-based lending protocol Term Finance lost approximately $8.5 million, roughly 68% of its vault product’s total value locked, in a single exploit. Blockchain security firms PeckShield and CertiK independently confirmed the loss, tracing approximately 2,843 ETH and 1.68 million USDC to an address initially funded through mixing protocol Tornado Cash.
What makes this incident significant is what did not fail. Term Finance’s Strategy Vaults were built on Yearn V3 architecture, audited, battle-tested vault infrastructure used across DeFi. The smart contracts performed exactly as written. Yearn confirmed publicly that the vulnerability existed in Term’s custom governance wrapper around the vaults, not in the base vault architecture itself.
Term’s governance design included specific safeguards against this type of attack. A seven-day timelock delayed governance transactions before execution. Vault liquidity providers (LP) could vote to veto queued transactions during that window. Neither mechanism prevented the exploit.
Term has not disclosed which governance role the attacker used or why the timelock and LP veto failed to stop the transaction. The takeaway does not depend on those details.
Audited code, governance timelocks, and community veto mechanisms were all in place. None of them mattered because the attack never touched the smart contract layer. It targeted the operational and governance infrastructure around the code. Code audits alone do not protect against attacks that bypass the code entirely. And increasingly, that is the direction attacks are moving.
How AI Is Accelerating Crypto Attacks in 2026
The Term Finance exploit was human-executed and governance-targeted. The broader threat landscape is shifting in a direction that makes these operational vulnerabilities significantly more dangerous: AI is reducing the cost and skill requirements of crypto crime at a pace the industry has not fully accounted for.
TRM Labs’ 2026 AI-in-Crime Adoption Index quantifies this shift. Overall AI adoption across crypto crime scored 54 out of 100, up from approximately 28 in 2024, a near-doubling in a single year. The adoption is not uniform. Scam operations have reached a “mature” level of AI usage, while hacking and ransomware remain at an “emerging” stage but are accelerating rapidly.
The scale of the acceleration is measurable. Reported losses from deepfake scams in 2026 have already surpassed the full-year 2025 total by 263%. The share of crypto scam reports involving AI-generated content, including deepfakes, AI chatbots, and synthetic identities, has increased by a factor of 13 since 2022.
In vulnerability discovery specifically, AI is changing both what gets found and how quickly. CertiK’s H1 2026 report identified 73 code vulnerability incidents targeting smart contracts that had been deployed for more than a year before being exploited, up from 45 for all of 2025. AI-assisted code analysis can examine significantly more contracts in a given period than human researchers, surfacing flaws in older, previously audited codebases that had been considered secure.
The structural shift is not that AI creates novel attack types. It is that AI removes the constraints that previously limited scale and accessibility. The skill floor has collapsed, the scale ceiling has lifted, and operations that previously required a team now require one individual with access to the right tools. Attacks that were previously uneconomical, such as targeting a wallet holding $20,000 or a mid-tier protocol with modest Total Value Locked (TVL), become viable when the marginal cost of each attempt approaches zero.
JadePuffer: The First Fully Autonomous AI Ransomware Attack
The AI adoption trends described above represent a gradual scaling of existing attack patterns. JadePuffer represents something qualitatively different.
JadePuffer is the first documented fully agentic ransomware operation, an attack where an AI agent executed the entire kill chain autonomously: reconnaissance, credential theft, lateral movement, privilege escalation, and encryption, with no human operator in the loop at any stage.
The significance for digital asset security lies in economics. A human-operated attack requires time, skill, and sustained attention per target, which naturally limits how many targets any given threat actor can pursue simultaneously. An autonomous AI agent removes that constraint. The same operation that previously required a skilled team spending days or weeks on a single target can now execute concurrently against hundreds of targets.
Blockaid’s H1 2026 report projected that AI agent deployments are growing approximately tenfold per year, and that the industry should anticipate multiple AI agent incidents in H2 2026, with prompt injection attacks as the leading vector, followed by tool-use abuse and unauthorized signing.
For institutions managing digital assets, the implication is straightforward: the window between vulnerability identification and exploitation is compressing. Defense models that depend on human response times, whether manual review of governance proposals, analyst-driven transaction monitoring, or alert-triggered incident response, face a structural speed disadvantage against autonomous agents operating at machine speed.
Why Infrastructure Attacks Dominate Institutional Losses
The macro data across individual incidents converges on a single conclusion. Infrastructure and operational compromises accounted for only 15% of incidents but drove roughly 76% of total losses. Financial damage is overwhelmingly concentrated in attacks targeting signing workflows, credential management, and governance mechanisms, not smart contract logic.
North Korea’s Lazarus Group was responsible for approximately 55% of all stolen funds in H1 2026. In two of the four largest incidents (Drift Protocol, $295M; KelpDAO, $293M), the attack vector was LinkedIn social engineering leading to multisig signer compromise, not a code exploit. Blockaid has noted there is no structural reason for this pattern to cease.
A finding from ack3’s research is perhaps the most instructive: across H1 2026, 94.4% of losses at audited victims fell outside the identified audit scope. Protocols that invested in code audits and passed them still lost funds, because the attack surface that produced the loss was never within the audit’s scope.
These are not three separate observations. They describe the same reality from three angles: the attack surface producing the largest financial losses in crypto is not the one that code audits, bug bounties, and smart contract reviews are designed to cover. It is the operational layer: governance wrappers, signing workflows, administrative credentials, and the human authorization step.
Why Audits and Code Reviews Are Necessary But Not Sufficient
Smart contract reviews, independent security audits, and bug bounty programs remain essential baseline security practices. The industry’s 74% reduction in DeFi exploit damage since 2022 is, in part, evidence that these measures are effective for the attack surface they are designed to cover.
The limitation is scope, not quality.
An audit verifies that code logic performs as specified under review conditions. It confirms that a smart contract executes according to its documentation. An audit does not, and structurally cannot, verify whether the governance wrapper around that code can be exploited, whether an administrator’s API token can bypass internal controls, whether a signing workflow permits a single compromised session to authorize a material transaction, or whether a governance timelock will actually prevent execution when triggered.
Term Finance’s vaults were built on audited Yearn V3 infrastructure. The audit was valid. The exploit succeeded because the vulnerability existed in a layer the audit was never designed to examine.
As AI compresses the interval between vulnerability discovery and exploitation, and as autonomous agents make simultaneous multi-target attacks economically viable, the gap between “audited code” and “secure infrastructure” becomes the primary exposure for any institution holding digital assets.
How to Defend Against Infrastructure-Level Crypto Attacks
If the attack surface has shifted from code to operations, defensive architecture must follow.
Two structural principles define effective institutional digital asset security in this environment.
The first is eliminating single points of failure in key management and signing. The Term Finance exploit, the Lazarus Group’s multisig signer compromises, and the broader pattern of credential-based infrastructure attacks share one characteristic: a single compromised component, whether a governance role, a signing key, or an administrative session, was sufficient to drain funds. Any custody architecture where a single device, individual, or credential can unilaterally authorize a material transaction carries the same structural risk, regardless of how many audits the surrounding code has passed.
The second is monitoring on-chain activity in real time, before settlement. Prevention remains the priority, but no prevention architecture is infallible. When prevention fails, the difference between a contained incident and a catastrophic loss depends on whether the unauthorized transaction is detected before it settles or during a subsequent review cycle. The 13.2% historical recovery rate for stolen digital assets, per SlowMist’s 2025 security incident data, makes the stakes of delayed detection clear: once funds have moved, the outcome is largely determined.
These two principles, structural prevention through distributed key architecture and real-time detection through transaction intelligence, are not alternatives. They are layers that must operate simultaneously, as each addresses the failure mode of the other. Leading institutional custody and KYT solutions have already embedded both into a single architecture, treating prevention and detection as one integrated system rather than two separate vendor relationships.
How ChainUp’s MPC Wallet and Compliance Stack Secures Institutional Workflows
Safeguarding enterprise assets requires an integrated approach that unifies distributed custody with real-time risk control. ChainUp bridges the gap between infrastructure protection and compliance oversight through a unified framework:
MPC Wallet: Eliminating Single Points of Failure
ChainUp’s non-custodial MPC wallet splits private keys into three independent shares. The full key is never assembled on any single device, server, or location, making single-party authorization structurally impossible.
- Three-layer security model: Encompasses the product layer (behavior monitoring, MFA, whitelisting, key recovery), system layer (risk control, alerts, audit trails, encryption), and wallet technology layer (MPC, multi-signature, HSMs with Intel SGX, key refresh), requiring independent defeat of each layer.
- Infrastructure-level policy engine: Enforces governance covering wallet member management, crypto transactions, whitelists, and custom risk strategies with configurable triggers, preventing unauthorized single-party actions where traditional timelocks and vetos fail.
- Tiered wallet strategy: Utilizes hot, warm, and cold wallets with configurable thresholds for risk exposure, asset percentage caps, transfer amounts, and real-time storage rules to balance institutional operational liquidity with deep cold storage security.
- Real-time transaction screening: Trustformer KYT embeds on-chain monitoring directly into the custody workflow, with 60ms detection speed and 20-layer deep fund flow tracing, so compliance operates at the same speed as the infrastructure it protects rather than as a separate review process.
Why the Integrated Architecture Matters
Custody without embedded compliance leaves a detection gap. Compliance without distributed key architecture leaves a prevention gap. In a threat environment defined by autonomous AI agents, industrialized social engineering, and governance-layer exploits, institutions require both operating as one system.
Across 300+ institutional clients, 200+ supported chains, and USD $300B+ in cumulative secure withdrawals, ChainUp’s integrated custody and compliance stack has maintained a zero-breach record, backed by SOC 2 Type II and ISO 27001/27017/27018 certification, and recognized by CNBC and Statista as one of the World’s Top Fintech Companies 2026 in the Digital Assets category.
Explore ChainUp MPC wallet Solution
