Bitcoin's flagship self-custody hardware wallet, Coldcard, lost roughly $116 million across more than 5,200 addresses in late July 2026. The root cause was a firmware bug from March 2021 that had quietly weakened seed generation on affected units for over four years. Anyone who set up a Coldcard between then and the patch is exposed.
The specific vulnerability aside, the incident is worth thinking about for what it says about the security landscape more broadly. Crypto protocols and their users now hold serious value. The tools attackers use to find and exploit vulnerabilities are getting sharper, and the class of tools now includes frontier LLMs. Several Chainflip liquidity providers have asked how we think about security at the protocol level, and what best practices apply to LPs. Here is our current view.
What Happened With the Coldcard Exploit
A concise recap of the facts.
Timeline. Attackers began sweeping addresses on July 30, 2026. Four waves followed. Roughly 1,816 BTC (~$116M) drained.
Root cause. Firmware v4.0.0, released March 2021, introduced an integration bug that routed seed generation to a software pseudorandom number generator instead of the STM32 hardware RNG (The Hacker News).
Effect. Effective seed entropy dropped from around 128 bits to as low as 40 bits. That is brute-forceable without physical access to the device.
Scale. Third-largest crypto hack of 2026. Year-to-date crypto losses now sit around $1.2 billion across 276 incidents (TRM Labs).
Coldcard's response. Halted shipments, destroyed unpatched units, and told all affected users to migrate to fresh seeds (Fortune).
What Made This Exploit Possible
Not attacker sophistication, just a quiet bug in trusted firmware, waiting five years to matter. Two takeaways for anyone thinking about their own setup:
Cold storage is not itself a security posture. It is a component. If the seed generation on that cold device is weak, cold storage of the seed protects you from nothing. The Coldcard incident is not about hot vs cold. It is about entropy at the moment of key generation.
Time does not fix latent bugs. Publication does. This exploit could have been discovered and disclosed responsibly at any point in the last five years. Instead, someone found it, kept quiet, and used it. That is the pattern to expect across the industry going forward.
The AI Dimension
Frontier LLMs are changing what a single attacker can do at scale. The evidence from the last twelve months of research is unambiguous.
Anthropic's smart-contracts research (Dec 2025) found that Claude Opus 4.5, Claude Sonnet 4.5, and GPT-5 collectively identified vulnerabilities worth roughly $4.6M in exploit revenue across their evaluation runs.
CyberChainBench, published in 2026, benchmarked the current generation of AI agents against smart-contract exploitation tasks. Codex with GPT-5.5 hit 43.7% on exploitation, realising $57.4M in total simulated exploit profit across 200 test cases.
Claude Opus 4.6, run via Claude Code, scored 45.6% on detection tasks per the EVMbench evaluation.
In one year, AI agents went from exploiting 2% of vulnerabilities in these evals to roughly 55%. Attack capability has scaled roughly 25 times.
Anthropic, OpenAI, and the wider frontier-lab community are pushing model capabilities forward on both sides of this equation. Anthropic's Claude and Fable families, OpenAI's ChatGPT and Codex, and the growing set of frontier models from other labs are getting better at finding vulnerabilities faster than human red teams, and symmetrically at helping defenders review code faster too.
For crypto specifically, the offensive advantage matters more in the short term. Public code is high signal for training and inference. On-chain balances are high signal for prioritising which targets to hit first. Every protocol with meaningful TVL is now an active LLM target, whether the team wants it or not.
What This Means for Chainflip LPs
Chainflip's design mitigates several failure modes that the Coldcard incident and the AI-attacker shift both touch:
Non-custodial protocol. Chainflip does not hold LP funds on a company balance sheet. Assets sit in the protocol's validator-secured custody model. No single vendor firmware bug can drain them.
Native settlement. Chainflip swaps settle in the native asset on the destination chain. There is no wrapped token to depeg, no bridge account to rugpull.
Utilization cap on lending pools. Even at high borrow demand, the protocol enforces a cap so there is always enough on hand to liquidate every open loan simultaneously. That protects both LPs and borrowers under stress.
Overcollateralization with continuous oracle price checks. Loans are checked against price feeds continuously. Liquidations happen at close-to-market rates via Chainflip's own on-chain DEX.
Two-year Boost track record. Over two years of Boost operation with no supplier deposits lost to Bitcoin reorgs or other issues.
These structural mitigations do not remove risk. They shift it. Where they leave LP-specific risk still open, the responsibility falls on the LP.
Practical Security Best Practices for Chainflip LPs
None of these are new. The Coldcard incident is a reminder that the basics still matter.
Wallet hygiene. Use a hardware wallet for any address holding meaningful assets, and verify the model's firmware version against the vendor's published security advisories on every major update.
Seed provenance. Prefer seeds generated on paper by dice rolls, or on multiple independent hardware devices, if you are seed-conscious enough to care. Never trust a single vendor's RNG unconditionally.
Small canary balances first. Fund a new wallet with a small amount, watch it for a week, then scale.
Separate hot and cold. Chainflip LP operations happen through an address that signs frequently. Keep the bulk of your treasury on a separate cold address that only signs to move funds to the hot wallet.
Monitor Chainflip Scan. Your positions are visible there. Set up alerts so you know when something moves.
Prefer widely-audited assets. Native BTC has one attack surface, the Bitcoin protocol itself. Wrapped variants add a custodian layer (Coinbase for cbBTC, BitGo for wBTC). None of these are "worse", but each adds a link in the chain of trust.
Update responsibly. When Chainflip ships new features or you rotate keys, run a test transaction with a small amount first, verify the destination address character by character, then commit larger volumes.
Assume your code review is not enough. LLMs are now good enough to find vulnerabilities in public smart contracts. If your LP setup involves any custom code (relayers, monitoring bots, scripts that automate signing), get it reviewed by a security professional, not just a colleague.
What We Do on the Protocol Side
Alongside the structural mitigations above:
All Chainflip protocol code is open source, and undergoes formal audits.
Active bug bounty program.
Continuous validator monitoring for anomalies and slashing conditions.
Public Chainflip Scan for LP position visibility.
Regular incident post-mortems when things go wrong (see the June 2026 incident report). We publish these because trust comes from transparency, not from claiming perfection.
Security Never Stops
Coldcard shipped one bug, quietly, five years ago. It cost $116M when the right person finally noticed. In 2026 that "right person" is increasingly an LLM running an unsupervised scan across the entire universe of public code and open protocols. Assume every layer of your stack is being probed.
For Chainflip LPs specifically: the protocol's non-custodial, native-settlement model removes some categories of risk that a Coldcard-style incident would touch. It does not remove the responsibility for basic wallet, key, and monitoring hygiene. Both matter.
If you have specific security questions about your Chainflip LP setup, reach out on Discord.
Resources
Swap - Start swapping native assets
Lending - Borrow against native Bitcoin
Blog - Product updates and announcements
Chainflip Scan - Track swaps and network activity
Website - Explore Chainflip
Earn with Chainflip:
Boost - Earn fees by providing single-sided liquidity with no IL risk
Stablecoin Strategies - Deposit stablecoins and earn optimized yields
Provide Liquidity - Supply assets to Chainflip's liquidity pools
Stake FLIP - Delegate FLIP and earn staking rewards
Find us:
