Over the past six months, the crypto industry confirmed 212 separate exploited incidents. Total losses: $1.1 billion. If you read only the dollar figure, H1 2026 looks like a slower year than H1 2025, when the Bybit heist alone inflated the stats. That would be a misread. The incident count is up 3.4 times year-over-year. The frequency curve is climbing faster than the loss curve, and that gap tells you more about where this industry is headed than any single headline.
Blockaid, the security firm that tracks much of this activity, is calling it a record incident count. I think it is a record of something deeper: the structural failure of an industry that still acts as if “audited” and “secure” are synonyms. They are not. They have not been for years. But the first half of 2026 made that unmistakable.
The raw data is brutal enough. 212 exploits. $1.1 billion in attacker proceeds. The top four incidents alone account for $707 million, or 64 percent of all losses. Operational security attacks — credential leaks, private key theft, signer infrastructure compromises, bridge backend intrusions — account for 74 percent of the total. North Korea-linked actors are responsible for 55 percent of all stolen funds. Those are the numbers that will make headlines. The ones that should make you nervous are buried underneath.
What Actually Happened
Let me calibrate the baseline. In H1 2025, the industry was looking at a headline total that was higher, largely because of the Bybit exploit. But if you peel that away, the number of incidents was far lower. What H1 2026 shows is not a single monster event; it is a swarm. Attackers are not aiming for one massive jackpot. They are running repeated, smaller strikes across dozens of protocols. The average incident size is smaller, but the ecosystem’s surface area is bleeding from every edge.
This shift matters because security teams and insurance underwriters build models on loss severity. A few giant events are easier to track, easier to attribute, and easier to prepare for. A swarming pattern is harder to defend against, because it rewards attackers who keep changing vectors. And that is exactly what we saw.
The technical story is not a story about smart contract vulnerabilities. For all the ink spilled on code audits, code flaws are not the main killer this half. The data is unambiguous: 74 percent of losses came from operational security failures. Private keys leaked. Credentials phished. Multi-sig signer laptops compromised. Backend servers cracked. Cross-chain bridge configurations botched. The smart contract code, in many cases, did exactly what it was designed to do. The problem was that attackers no longer need to break code when they can break the humans and machines holding the keys.
Deconstructing the Attack Vectors
Let me walk through the attack surface the way I would in a forensic review, starting with the largest categories and then moving to the emerging threats.
Operational Security: The 74 Percent Problem
Operational security attacks are not glamorous. There is no zero-day exploit to dissect, no clever flash-loan construction to trace through a dozen contracts. Instead, there is a CFO who clicked a phishing link, a developer whose GitHub token was left in a public repo, a validator key stored on a server with no hardware security module, a multi-sig signer who got socially engineered out of their seed phrase.
The Blockaid data shows that operational security incidents are fewer in count but much larger in average ticket size. That matches what I have seen in my own audits. One compromised key can drain a protocol’s entire treasury. A contract bug, by contrast, is usually limited to the value trapped in that one contract. Smart contract exploits are constrained by the size of the pool. Key exploits are constrained by nothing.
I have spent the better part of a decade on the operational side of this market. I’ve watched teams with formally verified contracts lose everything because a multi-sig signer’s laptop was compromised. That experience keeps resurfacing when I read this year’s data. The smart contract audit is no longer the perimeter. The perimeter is the key management pipeline.
Ethereum vs. Solana: A Tale of Two Attack Profiles
The loss attribution across chains is one of the most revealing details in the report. Ethereum-based projects lost around $332 million, and the majority of that came from smart contract vulnerabilities. Solana, on the other hand, saw over 98 percent of its losses attributed to private key and signature infrastructure being compromised.
That split is not random. It is a consequence of ecosystem design.
Ethereum is a matured settlement layer with a dense DeFi composability stack. Attackers target contract logic there because that is where Ethereum’s value has historically been concentrated: in complex, composable applications. Solana, by contrast, has grown so fast that the tooling and wallet infrastructure could not keep pace. Developers shipping on Solana have often prioritized speed-to-market over key custody design. In a high-throughput environment where everyone is racing to launch, private keys end up in environment variables, in CI pipelines, in not much better than text files.
From my seat at the exchange, I have watched support tickets spike every time a Solana mint or a new DeFi protocol goes live. The first question users ask is “Is my asset safe?” The second is “Should I pull liquidity?” The underlying fear is existential. The Solana data should force every builder on that chain to ask whether their internal security culture has grown as fast as their token price.
KelpDAO: The Single-Verifier Trap
The KelpDAO incident, at $292 million, is the single largest loss in the report. The LayerZero attribution is the key detail: the cross-chain message was forged because the bridge relied on what looked like a multi-sig setup but was effectively a single verifier configuration. That is not a bug in the usual sense. It is a governance failure. It is the kind of failure that no line of audit findings will catch, because the code is doing what it was instructed to do. The instruction itself was the flaw.
This is the “nominal multi-sig, actual single-verifier” structural defect. It is more dangerous than a code bug because it hides inside the trust assumptions that everyone takes for granted. The entire premise of cross-chain bridging is that no single entity should have the power to move funds. When the configuration collapses that power into one verifier, the building may still have fire exits, but the doors are all chained from the inside.
Call me paranoid, but I’ve seen this movie before. The pattern repeats across every bridge incident I’ve investigated: a team decentralizes on paper, centralizes in implementation, and then pats itself on the back for having a multi-sig. The audit report says “verified.” The attacker says “thanks.”
Drift Protocol and the Human Factor
Drift Protocol, at $285 million, shares a threat cluster with KelpDAO and Humanity Protocol, according to Blockaid. The attribution points to a North Korea-linked group, and the attack path is both sophisticated and depressingly basic. The compromise involved LinkedIn social engineering, targeted penetration, and multi-sig signer compromise.
The public reporting describes a six-month espionage operation. Six months. The attacker did not find a vulnerability in Drift’s code. They found a person. They studied that person’s professional network, their habits, their likely passwords, their trust in a fake recruiter or a fake collaborator. Then they converted that trust into a key.
This is the part of the security stack that most protocols still refuse to fund. They will pay $2 million for a smart contract audit and $0 for a mandatory hardware wallet policy, $0 for a strict signer isolation protocol, $0 for background checks on keyholders. The math is broken. The audit covers one snapshot in time. The attacker had six months to study the human beings who hold the keys.
I don’t trust audits anymore. I trust runtime monitoring. I trust key rotation schedules. I trust physical security for signers and cold storage for seeds. Audits are a point-in-time photograph. Attackers are a feature-film director. They have a script, a cast, and a production schedule. If you are not thinking in those terms, you are the B-story.
The New Vectors: AI Agents and EIP-7702
The report also flags the first recorded attack involving an AI agent. In the Bankr case, the attacker manipulated an AI agent into approving an unauthorized transaction. The loss was only $216,000. That is small change in the context of a $1.1 billion half-year. But the vector is brand new, and the speed at which it emerged should worry anyone who has been in this industry for more than one cycle.
AI agents are being deployed to automate treasury management, rebalancing, trading execution, and even governance participation. They are given permissions to sign transactions. They are given access to private APIs. They are, in effect, becoming new keyholders. And they do not have the threat awareness of a human. They have the threat awareness of their training data and their prompt context. Attackers are already learning how to manipulate both.
The EIP-7702 wallet delegation abuse is another early warning. Account abstraction, which is designed to give users flexible control over their wallets, also gives attackers a new surface to attack. If a wallet is delegated to a malicious smart contract, the user’s assets are at risk. The report shows that these vectors are not yet producing giant losses. But they are producing a rapidly expanding attack surface. The industry’s security model is already struggling to keep up with bridges and multi-sigs. Now it has to keep up with agents that can be manipulated, and wallets that can be reprogrammed.
I don’t do sentiment. I do infrastructure. And the infrastructure of account abstraction is beautiful until it is weaponized. The early numbers are small. The curve is not.
The Systemic Story Behind the Headlines
If you step back from the individual incidents, there is a more uncomfortable pattern. The industry is not losing $1.1 billion because attackers are getting smarter. It is losing $1.1 billion because protocols are handing over the keys and then calling it a hack.
The word “exploit” implies an attacker broke a lock. But when a private key is left on a server with no encryption, when a signer’s laptop is infected for six months, when a cross-chain bridge is configured with a single verifier, that is not a lock being broken. That is the door being left open. Calling it an exploit is generous. It shifts blame away from operational sloppiness and toward the collective mystery of “cyber attack.”
The data supports this reframing. The four largest incidents account for 64 percent of all losses. Those are not scattered accidents. They are concentrated, targeted failures of key management, human security, and infrastructure configuration. The industry can publish a thousand audit reports and it will not fix any of those four failures.
There is also a market dimension to this. Security incidents raise the risk premium for DeFi liquidity. When a major protocol gets drained, liquidity providers do not immediately run to the next high-APY farm. They wait. They demand higher yields to compensate for the risk. That means every exploit makes it more expensive for every legitimate protocol to attract capital. In a market where survival matters more than upside, this is a slow bleed.
The stolen assets themselves become another market force. $1.1 billion is a lot of capital. If even a fraction of that is laundered and sold, it becomes sell pressure on ETH, BTC, and the tokens of affected protocols. This is not a prediction. It is a plumbing reality. Hackers do not steal crypto to hold it. They steal it to move it, convert it, and fund operations. The market absorbs that flow, and the eventual distribution of these funds is rarely neutral.
The Contrarian Read: The 55 Percent Is Not the Real Threat
Here is the contrarian angle I keep coming back to. The report emphasizes that North Korea-linked actors are responsible for 55 percent of stolen funds. That is true, and it is a useful data point for sanctions enforcement and threat intelligence. But I would argue that the 45 percent from everyone else is the more serious signal.
The 55 percent is concentrated, state-backed, and relatively well understood. That is not to diminish it. But a concentrated attacker is easier to track, easier to model, and easier to defend against. The 45 percent is an amorphous, dispersed collection of opportunistic criminals, former insiders, fake support agents, ransomware gangs, and technical sociopaths. They do not share a command structure. They are not trying to fund a state program. They are just trying to get paid. That makes their behavior far harder to predict and far harder to stop.
If the industry over-indexes on North Korea attribution, it will build the wrong defenses. It will invest in geopolitical threat modeling and ignore the much larger problem of credential hygiene. The 74 percent operational security number applies across all actors, not just the state-sponsored ones. Every attacker, from the most sophisticated intelligence agency to the most amateur Discord scammer, benefits from a protocol’s weak key management.
There is also a hidden positive in the data: the Stellar Blend case, where tracking and attribution helped isolate $7.3 million. That is the first sign that the defense is starting to move from passive analysis to active response. Real-time monitoring, attack attribution, and rapid freezing are becoming the new security toolkit. The industry has spent a decade building firewalls around code. The next decade will be about building response pipelines after the code fails.
What the Market Should Be Watching
I have three forward-looking markers I will be watching in H2 2026.
First, watch the incident count, not just the dollar total. If the number of incidents continues to climb while the average loss stays small, the attack pattern is now firmly in swarm mode. That requires a different defense posture, one that centralizes threat intelligence and decentralizes key custody.
Second, watch the percentage of losses from operational security. If it stays above 70 percent, the industry is still failing to treat key management as first-class infrastructure. That number should be the KPI for the entire DeFi ecosystem.
Third, watch the AI agent experiments. The Bankr incident is a preview. Every protocol that gives an AI agent signing authority without a human-in-the-loop override is running a future incident. The industry is about to learn that prompt injection is the new SQL injection. It is only a matter of time before the losses from AI-agent manipulation stop being measured in six-figure losses and start being measured in eight.
The takeaway is not that crypto is broken. The takeaway is that the threat model has changed, and the security response has not caught up. The tools that worked in 2020 — smart contract audits, bug bounties, code reviews — are necessary but no longer sufficient. The protocols that survive this cycle will be the ones that treat key management like a national security issue, signer hardware like a bank vault, and human security like the real attack surface it has become.
Speed without security is fatal. The market rewards speed, but it punishes recklessness. H1 2026 was a reckoning. The question is whether anyone is actually listening.
Risk Warning
This analysis is not financial advice. Crypto assets are volatile, and protocol losses can cascade across interconnected markets. The protocol mentioned in this report may face legal claims, liquidity withdrawals, and civil or criminal investigations. Always do your own research and never deploy funds you cannot afford to lose. The author holds no positions in the protocols named in this article.
