Chainalysis pegs 2025 crypto scam inflows at $14B as attackers shift to permissions
Binance Research tied about two-thirds of April 2026’s $621M DeFi losses to access-control failures, not code bugs.
Chainalysis estimates on-chain inflows to crypto scams hit at least $14 billion in 2025, with scam clusters linked on-chain to AI vendors averaging far larger hauls per operation. Binance’s security team and research arm are pointing to the same pivot on the exploit side: attackers increasingly win by compromising approvals, credentials, and governance controls rather than finding smart-contract bugs.
Key Takeaways
- On-chain inflows to crypto scams totaled at least $14 billion in 2025, based on Chainalysis estimates.
- Scam operations with on-chain links to AI vendors averaged $3.2 million per operation versus $719,000 for those without, a relationship Chainalysis described as correlation rather than causation.
- Access-control failures drove roughly two-thirds of April 2026’s $621 million in DeFi exploit losses, per Binance Research.
- Binance CSO Jimmy Su said attackers are moving from contract exploits to people, credentials, and governance systems, citing a $1.2 million governance attack on BrainTrust that Binance Security helped prevent.
Chainalysis Puts 2025 Scam Inflows at $14B as AI-Linked Ops Show Bigger Hauls
Chainalysis’ estimate of at least $14 billion in 2025 on-chain inflows to crypto scams is a scale marker, not a complete accounting. “On-chain inflows” here means funds that can be observed moving on public blockchains into addresses associated with scams, which captures what hits known destinations and misses what never gets labeled or gets obscured.
Inside that aggregate, Chainalysis flagged a subset of scam operations with on-chain links to AI vendors that averaged $3.2 million per operation, versus $719,000 for scams without those links. The firm did not frame that gap as proof that AI caused higher proceeds. It is a correlation, and the packet does not include a breakdown of what portion of the $14 billion is specifically AI-enabled.
The mechanism Chainalysis associates with AI-linked scam clusters is not exotic contract engineering. It is identity fabrication at scale, using deepfakes, face-swapping, and large language models to impersonate real people and build credible fronts quickly.
From Code Bugs to Clicks: Why Authorization Layers Are Becoming the Primary Failure Point
The cleanest way to steal from a protocol is still to make the protocol authorize the theft. That is why the attack surface is drifting up the stack from Solidity edge cases to the systems that decide who is allowed to sign, upgrade, pause, or move treasury funds.
Binance Chief Security Officer Jimmy Su put the shift bluntly: “Code is no longer necessarily the weakest link in Web3. As smart contract security improves, attackers are shifting their attention to the people, credentials and governance systems surrounding protocols,” Su said. He cited Binance Security helping prevent a $1.2 million governance attack on BrainTrust, and argued that defending a protocol now means securing “not just its code, but also who can control it, how that control is exercised, and the infrastructure and people behind it.”
That framing matters for traders because governance and permissioning risk is not theoretical. A “governance attack” is an attempt to manipulate a protocol’s decision-making or control mechanisms, including votes, admin keys, or role-based permissions, to steal funds or change rules. If the attacker can get a privileged signer, a compromised admin key, or a coerced upgrade through, the contract can behave exactly as designed while still producing a loss.
The social layer is where AI changes the economics. Fractl founder and CEO Rime Salmi described the investor-impersonation dynamic as a volume business: “There are a lot of people who are posing as investors and they're absolutely not. They're looking for investments or grifting actually.” AlongID CEO and co-founder Erika Maslauskaite made the same point from the identity side: “With AI currently, you can fake everything. You can literally fake everything,” she said, framing the core question as “how do we verify what is true and authentic and what is not.”
Access-Control Failures Dominate a Month of DeFi Losses, Binance Research Says
Binance Research’s April 2026 loss breakdown puts a number on the permissions thesis. In that month alone, access-control failures accounted for roughly two-thirds of the $621 million lost to DeFi exploits, according to the report.
An “access-control failure” is not a single bug class. It is a breakdown in who is allowed to do what, across keys, roles, and operational processes, that lets an attacker execute actions without needing a novel contract vulnerability. In practice, that can look like compromised admin credentials, misconfigured role-based access, or governance pathways that can be manipulated faster than defenders can react.
Wallet UX is where this becomes operational for power users. Para founder and CEO Nitya Subramanian described wallets as the choke point for on-chain action: “Wallets are ultimately the authorization and control flow layer of anything that's happening on chain. Every chain, every DeFi primitive, every action that you can take on chain needs to go through a wallet. And I feel like people still don't fully get that.”
The catch is that the same transparency that makes on-chain systems auditable does not automatically solve attribution. BitOK founder and CEO Dmitry Machikhin said mixing remains a persistent obstacle: “Even Chainalysis has no, like, 100% solution for the mixing.” Mixing refers to techniques and services that obfuscate the origin of crypto by pooling and redistributing funds to make tracing harder.
Machikhin also argued that enforcement signals do not always propagate into effective blocking. “There was a case where the Israel government seized some wallets in their official documents… But this didn't happen,” he said, referring to the expectation that seized addresses should be blocked by exchanges and other entities. His summary of tracing limits was even more direct: “We are not catching anybody. We are just showing … the path of money.”
AI Agents Raise the Stakes for Identity: On-Chain Registries vs ‘No Transaction Layer Yet’
The forward-looking risk is that “who signed this” becomes harder to answer as software starts signing for people. Concordium chief growth officer Varun Kabra said AI agents transacting on users’ behalf are “already starting,” describing a near-term path where agents “pay for things,” “sign up for services,” and “probably handle your financial transactions now.” He put agent traffic “six to twelve months away” from overtaking human transactions, and called “human to agent accountability … the biggest problem I think the world needs to solve for.” That timeline is a claim, not a measured trendline in this packet.
Concordium’s proposed mitigation is identity plumbing onchain. The company said its Agent Registry had 1,131 agents and more than 15,663 on-chain transactions as of July 14, 2026, and described tying agent identities to verified human owners using selective disclosure and zero-knowledge proofs. Selective disclosure means revealing only the minimum necessary identity details for verification. A zero-knowledge proof is a cryptographic method to prove something is true without revealing the underlying private information.
There is a hard counterpoint from the infrastructure side. XDC Network co-founder Atul Khekade said: “AI doesn't have a transaction layer right now.” He added: “AI platforms don't have a monetization compliance layer that they can use for, like, real transactions to execute actions,” arguing counterparties cannot build that missing layer “overnight.”
For traders, the near-term tells are measurable. The missing breakdown is what portion of Chainalysis’ $14 billion scam inflows is actually AI-enabled, and whether future updates quantify it. On the exploit side, new monthly loss reports will either confirm April 2026 as a permissions-heavy pattern or mark it as a noisy outlier. The other concrete signal is whether more case studies emerge like the prevented $1.2 million BrainTrust governance attack, especially those involving admin keys, role-based access, or vote manipulation.
My Read: The Next ‘Exploit Season’ May Look Like Social Engineering Plus Permissions, Not Reentrancy
The part that decides this cycle is not whether smart contracts are “secure.” It is whether the authorization perimeter around them is getting tighter faster than attackers can scale impersonation, and Binance Research’s April 2026 breakdown suggests it is not an edge case when two-thirds of a $621 million loss month is tied to access control.
Chainalysis’ $14 billion minimum for 2025 scam inflows sets the baseline, and the $3.2 million versus $719,000 per-operation gap in AI-linked clusters is the uncomfortable detail. If future reporting can quantify how much of that $14 billion is actually AI-enabled, and monthly exploit data keeps pointing to access-control failures, the security pivot becomes a structural risk to governance and wallet approval flows rather than a narrative about smarter hackers.
