After two quiet months, July 2026 broke the pattern hard. Roughly $247.4 million was stolen across the month — the worst total since April’s record $644 million, and more than triple June’s $75.87 million.
But the headline number is not the story. The composition is.
24 DeFi protocols lost a combined $132.2 million in July — an ordinary-to-heavy month by 2026 standards, spread across two dozen separate incidents, each with its own audit report, its own post-mortem, its own root cause.
One hardware wallet flaw lost more than all of them. The Coldcard entropy vulnerability accounted for roughly $110 million and climbing by month-end, from a single bug, in a single product, introduced in a single build five years earlier.
For the first time in this industry’s history, the largest loss vector in a month was not a smart contract, not an exchange, not a bridge, and not a phishing campaign. It was cold storage.
The Month at a Glance
| Category | Approx. losses | Incidents |
|---|---|---|
| Hardware wallet (Coldcard entropy flaw) | ~$110M+ | 1 root cause, 5,200+ addresses |
| DeFi protocol exploits | ~$132.2M | 24 |
| Everything else | remainder to ~$247.4M | — |
| Total | ~$247.4M | Second-worst month of 2026 |
The Coldcard Event
We published a full technical breakdown of the Coldcard failure on 3 August. The short version, because it reframes how the rest of this recap should be read:
A March 2021 firmware build error routed Bitcoin seed generation through MicroPython’s software pseudorandom number generator instead of the device’s hardware TRNG. Effective seed entropy fell from 128 bits to roughly 40 bits on older models — brute-forceable remotely, with no physical access to the device required.
On 30 July, attackers drained approximately 1,083 BTC ($70.2 million) from 1,196 addresses in 41 minutes — about 30 hours before Coinkite’s public advisory. Subsequent waves pushed the total past 1,816 BTC / $116 million from 5,200+ addresses, with recovery ongoing into August.
Three things make this categorically different from every other entry in this recap:
- The victims did nothing wrong. No phishing click, no malicious approval, no key exposure. The air gap held. The key was weak at creation.
- It cannot be patched for the affected. New firmware protects future seeds. It cannot retroactively add entropy to a seed generated in 2022.
- The losses are permanent and ongoing. Any affected seed that still holds a balance remains in the attacker’s search space indefinitely.
The DeFi Half
Underneath the hardware story, July’s 24 DeFi incidents followed the year’s now-familiar distribution. The dominant categories:
- Privileged access abuse — compromised admin keys, multisig signer compromise, and pre-signed transaction manipulation. Still the leading cause by incident count, as it has been since May.
- Oracle and pricing manipulation — flash-loan-assisted price distortion against protocols with thin or single-source collateral pricing.
- Bridge and cross-chain messaging failures — the perennial. Every quarter of 2026 has produced at least one.
- Infrastructure compromise — cloud credentials, CI/CD access, and DNS, sitting entirely outside the scope of any smart contract audit.
The absence in that list matters as much as the presence. Pure Solidity logic bugs — reentrancy, arithmetic errors, unchecked calls — are no longer where the money goes. A decade of tooling, better languages, standard libraries, and audit practice genuinely worked on that category. Attackers moved.
Two Trends Worth Naming
1. The attack surface has moved out of the contract and into everything around it.
2026’s cumulative losses now break down overwhelmingly into human, operational, and supply-chain failures: social-engineered signers at Drift, a leaked cloud key at Resolv, an admin key compromise at Wasabi, a bridge boundary at Aztec, and now a firmware build flag at Coinkite.
Every one of those systems had been audited. None of the audits covered the thing that failed, because the thing that failed was not in scope. The industry’s assurance model is aimed at the smallest remaining part of its attack surface.
2. Self-custody now carries a quantified tail risk.
The custody debate has always been framed as a binary: trust an exchange (Mt. Gox, FTX, Celsius) or trust yourself. July supplied the missing data point — self-custody has a vendor-concentration tail too, and it can materialise at nine-figure scale, simultaneously, across thousands of users who each believed they held independent risk.
They didn’t. Five thousand people who each thought they had their own key had, in effect, shares in one firmware build.
The structural answer is the one that has always been available and rarely used: multisig across hardware from different vendors. A 2-of-3 across three manufacturers means a single vendor’s entropy failure costs one key, not the coins. In July, that would have been the difference between an inconvenience and a total loss.
Where 2026 Stands
| Period | Approx. losses |
|---|---|
| H1 2026 (Jan–Jun) | ~$750M |
| July 2026 | ~$247.4M |
| YTD through July | ~$1B |
Roughly a billion dollars, seven months in, with April and July supplying more than 85% of it between them. The baseline months run $70–80 million; the catastrophic months run $250–650 million and arrive without warning.
Any security budget built around the average is built around a number that has never once occurred.
Checklist Coming Out of July
For self-custody users:
- Audit which device and firmware version generated each of your seeds. If you cannot answer that question for a wallet holding meaningful value, migrate it.
- Move to multi-vendor multisig for anything you would be devastated to lose.
- Use dice or coin-flip entropy input where your device supports it.
For protocol teams:
- Inventory every privileged role and every credential outside your contracts — cloud, CI/CD, DNS, package registries. That inventory is your real attack surface.
- Add supply-chain and dependency provenance to your assurance programme. The Coldcard flaw was a build configuration error; yours might be too.
- Establish a relationship with an on-chain monitoring firm before you need one. Response time is measured in minutes.
August has already opened with a payment processor drained and its proceeds converted straight to Monero. More on that shortly.
Sources: Cointelegraph · Coinmonks · BingX · CCN
