Infrastructure, not ambition, decides outcomes.
Blockchain, cybersecurity, and artificial intelligence are converging on the same fault line: organizations adopt the technology faster than they build the infrastructure to hold it. This whitepaper lays out the gaps — in current, sourced figures — the named incidents behind them, and what closing them looks like.
The largest theft in the history of crypto did not break a blockchain. In October 2026, roughly $6.3 billion left the world's largest exchange in about twenty minutes (TRM Labs; ~$3B since frozen) — and in every major incident before it, the failing component was human infrastructure: a compromised developer machine, a signer's laptop, a portal without multi-factor authentication.
Meanwhile the legitimate side compounds: on-chain tokenized assets doubled in a year to $39.4B, with BlackRock, Franklin Templeton, and Circle each running billion-dollar tokenized funds — and enterprise AI adoption is moving from under 5% of applications to a projected 40% in a single year. The gap between what companies adopt and what they can securely operate is the defining risk of this cycle — and the subject of this paper.
Capability acquired faster than the controls to operate it is the common denominator of every incident in this paper.
The blockchains held. The signing infrastructure didn't.
The record thefts share one pattern: the smart contracts were not the weak point. At Bybit (February 2025, $1.46B in ETH — the record until October), attackers compromised a developer machine at wallet platform Safe and served three multisig signers a doctored interface; each confirmed a "normal transfer" on a hardware wallet that was actually replacing the wallet's implementation contract (FBI attribution: DPRK's TraderTraitor/Lazarus).
At Radiant Capital (October 2024, ~$50M), malware on the signers' computers made a malicious transaction display as benign on their Ledgers — the Safe contract itself was untouched. At DMM Bitcoin (May 2024, ~$305M), the attackers never touched the exchange: they hit its wallet software vendor through a fake recruiter message. And Bitget lost $351.6M in September 2026 to a backend compromise.
The trend line is unambiguous. CertiK counted ~$2.45B stolen across Web3 in 2024 and ~$3.35B across 630 incidents in 2025; Chainalysis recorded $3.4B in 2025 alone, with private-key/signing compromises the dominant vector (43.8% of 2024 losses), and 158,000 individual wallet compromises in 2025 — nearly triple 2022.
Every record theft of this cycle was a signing-operations failure, not a cryptography failure.
What signing discipline looks like when written down — the kind of policy every treasury should be able to produce on request:
# treasury signing policy — example quorum: 3-of-5 signers: - role: ceo device: ledger-a # offline, geographically separate - role: cfo device: ledger-b # offline - role: ops-lead device: ledger-c - role: counsel device: ledger-d - role: backup custody: sealed-envelope # ceremony-governed rules: - movements > 100k USD: 24h timelock + notice to all parties - destination changes: verified on an independent, offline decoder - signer screens: never the sole source of transaction truth verification: - decoded calldata checked against the human intent, off-device - destination allow-list maintained outside the signing stack
Non-custodial and multi-signature wallet architecture run as an operation, not a one-time setup: signer device hygiene, transaction-visibility so every party can verify what is actually being signed, key ceremonies, and systems audits covering the signing chain end to end. See Blockchain & Web3 and Cybersecurity.
The infrastructure isn't.
Distributed on-chain RWA value went from ~$18.6B (November 2025) to ~$39.4B (November 2026). Tokenized US Treasury funds alone hold ~$15.7B. The largest asset managers now run tokenized products at scale: Circle's USYC (~$2.39B), Ondo's USDY (~$2.36B), BlackRock's BUIDL (~$2.17B, nine chains), and Franklin Templeton's BENJI (~$1.71B) — figures as of October 2026.
The forecasts explain why institutions keep entering despite the thefts: McKinsey's base case is ~$2T of tokenized securities by 2030; Standard Chartered projects $30.1T by 2034. An eight-fold spread between credible forecasters is itself the signal — the variables are regulatory clarity and custody infrastructure, not demand.
And much of what has been tokenized sits inert: about 56% of reported RWA value is idle — on-chain but not productively deployed (thirdweb, 2026). Issuance without operational infrastructure (compliance monitoring, reporting, custody that survives jurisdictions) is where tokenization projects stall.
Demand is not the bottleneck. Issuers are discovering that the token is the smallest part of the system.
We've tokenized physical assets — gold, oil drilling equipment and production — across jurisdictions, including cross-jurisdictional compliant custody and the full issuance stack: token, contracts, presale, cap table, vesting, and ongoing operations. See Blockchain & Web3.
One stolen password, no MFA, $3.09 billion.
The clearest case study in what "unmanaged infrastructure" costs: in February 2024, attackers logged into a Change Healthcare remote-access portal with stolen credentials. The portal had no multi-factor authentication. Nine days later ransomware took a core piece of U.S. healthcare infrastructure offline for weeks; the final count reached 192.7 million people, the largest healthcare breach ever recorded, with ~$3.09B in costs and ~$9B advanced to providers whose claims stopped processing.
Zoom out and the denominator is staggering: global cybercrime is estimated at $10.5 trillion for 2025, up from $3T in 2015 — third-largest economy on earth if measured as a country (Cybersecurity Ventures). Most of that lands on organizations with unmanaged servers, unaudited access control, and no incident readiness — the same profile as the crypto victims above, minus the blockchain.
The Change Healthcare breach began with one login. The Bybit theft began with one developer laptop. The pattern is identical.
Managed hosting and operations on bare metal and cloud; systems audits covering applications and contracts, server and network configuration, access control, and incident readiness — ending in a formal report with severity rankings, a remediation plan, and a re-test; hardening executed by us or documented for your engineers. See Cybersecurity.
From 5% to 40% of enterprise apps in one year.
Gartner's 2026 CIO survey found only ~17% of organizations have deployed agents, while over 60% expect to within two years — the steepest adoption curve it tracks. The agentic AI market sits near $10B in 2026 with analysts projecting 40%+ compound growth for the decade (Grand View Research; MarketsandMarkets); Gartner's best case has agents driving ~30% of enterprise software revenue by 2035.
The capability gap is the opportunity and the risk: agents that can act — move data, call APIs, approve workflows — inherit the same failure modes as the sections above, now with machine speed. Deployed without operational discipline, agentic systems become a new, faster signing infrastructure to compromise.
An agent with credentials is a signer. Treat it like one.
Agent deployment done right starts from a permission scope — written before the agent is:
# agent permission scope — example agents: treasury_monitor: read: [balances, positions, oracle_feeds] write: [] # observes; never mutates alert: compliance_channel payment_proposer: read: [invoices, approvals] write: [] # proposes only propose_to: multisig # 3-of-5 human quorum signs spend_limit: 0 # no unilateral authority, ever ops_runner: read: [logs, metrics] write: [deploy_pending] # gated behind human approval # rule: an agent may prepare an action; only a quorum executes it
Agents deployed into operations — built on open-source systems like Claude Code, Hermes, and OpenClaw, or closed-source where requirements demand — plus the machine learning behind forecasting, detection, and document processing. See Artificial Intelligence.
Every gap in this paper is an operational-controls gap.
A compromised signer emptied Bybit. A stolen password without MFA cost Change Healthcare $3.09B. Idle tokenized assets and ungoverned agents are quieter versions of the same defect: capability acquired faster than the controls to operate it. Attackers have noticed — threat intelligence through 2026 shows state-affiliated groups running long-horizon social engineering against exactly these seams, and increasingly using AI to do it.
The organizations that come through this cycle will be the ones that treat custody, security, compliance, and AI as one operational discipline — researched before built, auditable by design, and run by people trained to operate it.
Custody, security, compliance, and AI are four names for one discipline: operating powerful systems without single points of failure.
Three ways this shows up.
The company holding crypto on a balance sheet
Treasury in crypto, but the setup — exchange accounts, one hardware wallet, one person with the seed phrase — won't survive an audit or a resignation. After Bybit, boards ask what happens when signers see a doctored screen. We replace it with multi-signature custody, signer policies, transaction visibility, board-readable reporting, and staff training.
The asset owner entering tokenization
Real assets — commodities, equipment, production — that need to go on-chain without inventing the machinery, and institutional counterparties who require compliant custody before they'll touch it. We advise on what to tokenize, build token and contracts, run issuance and burning, structure cross-jurisdictional custody and compliance, and operate it with you.
The organization modernizing with AI
Competitors deploy agents while your team reads about them. We assess where agents and models genuinely pay off, deploy them — open-source or custom — with the same signing discipline as the treasury, and train your people to run and trust what was built.
Blockchain & Web3
→Wallet and smart-contract management · asset tokenization · governance · custodial and non-custodial systems · compliance monitoring · reporting & risk frameworks.
02Cybersecurity
→Server management · systems audit · hardening.
03Artificial Intelligence
→Agentic systems · machine learning.
If one of these gaps is yours, book a fifteen-minute discovery call — direct, no pitch team.
- TRM Labs / Chainalysis, October 2026 Binance theft reporting (~$6.3B outflow; ~$3B frozen).
- FBI public advisory, February 2025 — Bybit theft attributed to DPRK TraderTraitor/Lazarus; CertiK Hack3d 2025 annual report (~$3.35B, 630 incidents) and H1 2026 review.
- Chainalysis, 2025 Crypto Crime Report ($2.2B stolen 2024; private-key compromises 43.8%) and 2026 crypto-hacking update ($3.4B in 2025; 158,000 wallet compromises).
- Radiant Capital post-incident reports (October 2024, signer-endpoint compromise); DMM Bitcoin / Ginco attribution (FBI, Japan NPA).
- RWA.xyz via State of RWA 2026 reports ($39.4B distributed on-chain RWA value, Nov 2026; tokenized Treasuries ~$15.7B); CCN/RWA.xyz fund sizes Oct 2026 (USYC, USDY, BUIDL, BENJI).
- McKinsey (2024), Standard Chartered (2024), BCG/Ripple (2025) tokenization forecasts; thirdweb (2026) idle-asset analysis.
- UnitedHealth Group earnings disclosures and HHS OCR notifications — Change Healthcare breach ($3.09B; 192.7M individuals); Privacy Rights Clearinghouse 2025 breach report.
- Cybersecurity Ventures, Official Cybercrime Report 2025 ($10.5T estimate).
- Gartner (Aug 2025 press release; 2026 CIO survey) — enterprise agentic AI integration and adoption; Grand View Research (Mar 2026); MarketsandMarkets (Apr 2025) — agentic AI market sizing.
Figures are third-party estimates and incident reports, cited with their sources and dates; early figures for ongoing incidents (including the October 2026 Binance theft) are preliminary and subject to revision.
