7 Key Points:
1/ Crypto lost over $2.2 billion (maybe more now) across 288 documented attacks by late September 2026, putting the year on pace to rival the worst on record.
2/ Off-chain attacks accounted for 76 percent of hack losses in the dataset. Private keys and smart contracts were often not the main weak link.
3/ The Trezor impersonation case cost one person $282 million after a fake support contact obtained a seed phrase.
4/ Coldcard lost $112.7 million through a firmware randomness flaw. Investigators said intruders used unrestricted AI models to find and exploit the weakness.
5/ Bitget lost $387.5 million after a third-party zero-day and stolen internal credentials let criminals push fraudulent withdrawals. Kelp DAO lost $292 million from a cross-chain bridge configuration with a single point of failure. Supply chain and configuration risks are central.
6/ AI is already cheap and scalable. CrowdStrike saw an 89 percent surge in AI-enabled threats and 88 percent of vulnerability attacks starting within 48 hours of a public proof of concept. Fortinet recorded a 389 percent rise in ransomware victims.
7/ Quantum planning matters, with experts warning of relevant machines in the 2029 to 2030s range, but it is a multi-year migration problem.
The numbers land on you like a heavy weight. By late September 2026, the crypto industry had already lost more than $2.2 billion across 288 documented attacks. That figure puts this year on pace to rival the worst on record. The headline number tells a misleading story if you stop there. The real story is in the data, and it points to a threat that looks nothing like the one most boardrooms and security teams worry about.
Take a breath and look at the shape of the damage. In January, a single person lost $282 million after an impersonator posing as Trezor support convinced them to hand over a seed phrase. That was not a broken smart contract. That was a human being trusting a voice on the other end of a message. In July, intruders drained $112.7 million from Coldcard hardware wallets by exploiting a firmware flaw that affected how those devices generated randomness. In September, Bitget lost $387.5 million after criminals exploited a zero-day in third-party security products and used stolen internal credentials to push fraudulent withdrawal commands through the exchange’s own systems. In April, Kelp DAO lost $292 million due to a cross-chain bridge configuration that created a single point of failure for bad actors to poison.
Now look at what those incidents have in common. Private keys were not the weak link in most of them. Code was not the only weak link either. The losses moved through trust, through supply chains, and through the tools and people around the code. Off-chain attacks, the kind that compromise credentials, manipulate people, and poison supply chains, accounted for 76 percent of hack losses in the data set behind that $2.2 billion figure. The threat has shifted away from vanilla smart contract exploits and toward the infrastructure and human layers that hold everything together.
That migration matters because it tells us where defense dollars and attention belong right now. It tells us something else, too. The most urgent fight is not the one against a future quantum computer. The most urgent fight is the one against attacks happening today, many of which use artificial intelligence that is already here and already cheap.
The economics explain the shift. A phishing kit costs almost nothing and reaches on thousands of targets in minutes. A compromised vendor account opens a path into every customer the vendor serves. A stolen session token can bypass the password and the second factor in one move. Security teams that still model the adversary as a lone coder hunting for a reentrancy bug are modeling a smaller threat than the one that actually shows up. The intruders who took $2.2 billion this year studied the humans, the vendors, and the tools. They did not wait for a physics breakthrough.
Consider the pace of change. Verizon’s 2026 Data Breach Investigations Report found that 31 percent of confirmed breaches involved exploitation of software vulnerabilities, and that vulnerability exploitation overtook stolen credentials as the primary initial access vector for the first time. CrowdStrike observed an 89 percent surge in AI-enabled cyber threats, with 88 percent of attacks targeting vulnerabilities occurring within 48 hours of a public proof-of-concept release. Fortinet recorded a 389 percent year-over-year increase in ransomware victims as criminals used AI to automate and scale their operations. These are not projections about some distant future. These are measurements of the present.
The Coldcard case shows how directly AI has changed the game. Investigators found that intruders used unrestricted AI models to uncover and exploit a firmware weakness. On the defensive side, Coinkite used frontier AI tools to audit its entire system after the breach and found problems unrelated to the original bug. AI is not waiting for quantum computing to arrive before reshaping the balance between attackers and defenders. It is reshaping that balance right now, and security teams that treat AI as an afterthought are losing ground.
Why does the quantum conversation keep pulling attention away from this? Partly because quantum feels like the ultimate threat, the kind of thing that justifies major strategic planning and long timelines. The Government Accountability Office reports that experts believe a cryptographically relevant quantum computer could arrive as soon as the 2030s. Google warned in March 2026 that one could arrive by 2029. Those warnings deserve attention. Cryptographic migration takes years, and the harvest-now-decrypt-later problem means some data already sits in intruders’ hands. Nobody should dismiss quantum planning as fantasy.
Planning for quantum and defending against today’s attacks are not the same job, and they do not compete for the same urgency. A quantum computer capable of breaking RSA-2048 does not yet exist. An AI-assisted phishing campaign that drains a nine-figure wallet exists today. A supply chain compromise that evades a third-party security product exists today. A social engineering operation that convinces a person to surrender the one secret that protects everything exists today. The losses are not theoretical. They sit on chain, block by block, month after month.
The instinct to talk about quantum first feels prudent, but it gets priorities backward. Quantum is a migration problem with a horizon measured in years. AI-powered attacks on current code and people are an operational problem with a horizon measured in hours. CrowdStrike found that 88 percent of vulnerability attacks began within 48 hours of a public proof of concept. A defender who spends this quarter building a post-quantum roadmap but does not harden the identity layer, audit the supply chain, and train people to recognize impersonation is preparing for the wrong fight at the wrong time.
This is not an argument against quantum readiness. It is an argument about sequence. First, get the current code and the current humans into a defensible position. Treat AI as the primary adversarial capability it already is, not as a curiosity to address after leaders sign off on the quantum migration plan. The 2026 data is screaming a clear message. The billions are not disappearing because someone built a quantum computer. They are disappearing because criminals are using available tools against the weaknesses in the systems and people we rely on right now.
Anndy Lian is an early blockchain adopter and experienced serial entrepreneur who is known for his work in the government sector. He is a best selling book author. The latest book is Web4: The Age of Autonomous Intelligence.
