Bitcoin’s security crowd is making a blunt ask: give vetted defenders access to frontier AI before attackers use it first.
- 40+ crypto groups want controlled AI access for trusted security researchers
- Kimsuky shows attackers are already using local AI tools
- Bitcoin security bugs keep surfacing, from wallet flaws to AI-assisted code review findings
- No major AI lab had publicly answered the coalition’s specific request at the time of writing
The Bitcoin Policy Institute said in an Aug. 10 X post that more than 40 digital-asset organizations are pushing leading AI labs to set up a controlled program for qualified open-source security researchers. This is not a call for public access. It asks for early model access, computing power, secure environments, and direct communication with lab security teams.
That distinction matters. This is not “open the floodgates and hope for the best.” It is a defensive program for vetted researchers who inspect public code, wallets, and infrastructure before attackers do.
The coalition’s logic is simple. AI-powered attacks are getting better, while current model safeguards can still slow down legitimate security work. At the same time, criminals and state-backed groups can use open-source models or run them locally, outside the usual cloud-based guardrails.
According to Reuters’ summary of the issue, the North Korea-linked hacking group Kimsuky reportedly used local AI environments built with Ollama, GPT4All, and Msty. Local setups like that can support malware development, data analysis, phishing, and attack automation. In plain English: if the bad guys can run models on their own hardware, platform-level safety filters only go so far.
A prior crypto.news report said Kimsuky also produced AI-generated phishing material aimed at cryptocurrency, investment, and fintech companies. Phishing is still the oldest scam in the internet playbook, just dressed up with newer tooling and a more convincing mask.
The request from Bitcoin defenders comes with a very practical backdrop. The ecosystem keeps tripping over real security problems.
One example is the Coldcard hardware wallet issue tied to entropy, the randomness used to generate seed phrases. Seed phrases are the backup words that can recover a wallet. If the randomness behind them is weak, the wallet may be easier to attack than users realize. That is not some abstract crypto lecture. It is the difference between a secure key and a lottery ticket handed to thieves.
Galaxy Research has updated its own view of the Coldcard-related losses over time. A newer summary says Galaxy later raised its observed estimate to 1, 367.05 BTC, worth about $88.6 million, across 4, 585 addresses. That newer figure appears to supersede earlier estimates cited in previous reporting, which is a good reminder that active security investigations can shift as the evidence changes.
In other words, estimates move, methodologies differ, and on-chain patterns do not always tell the full story. They can point to a likely operator or a cluster of addresses, but they do not magically prove every wallet in sight was affected in exactly the same way.
Then there is the AI-assisted review work itself. Bitcoin Red Team reported 4, 962 potential issues across 390 Bitcoin-related projects in fewer than 30 hours of AI-assisted code review. Of those findings, 720 were initially classified as high or critical severity. More than one-fifth had been reproduced when crypto.news covered the effort.
That last detail is the important one. Huge output numbers sound impressive, but raw findings are not the same thing as confirmed bugs. AI can help triage code, surface suspicious patterns, and prioritize human review. It can also produce noise, duplicates, and false positives at industrial scale. Security theater at machine speed is still theater.
Still, the broader point holds: defenders need better tools, and they need them before attackers do. If the offense can use increasingly capable models to generate phishing, automate reconnaissance, and test exploits, then locking researchers out of the best defensive tooling is a weird way to run a security regime.
The industry is also putting real money behind the problem. The Bitcoin Security Consortium was recently established by Strategy, BlackRock, Coinbase, and six other companies, with a pledge of $15 million over three years. Its initial focus is post-quantum cryptography, meaning cryptographic methods designed to resist future quantum computers.
That is not a doomsday declaration. It is long-range planning. Serious institutions are preparing now because security migrations take time, and waiting until the threat is immediate is how you end up doing emergency surgery with a butter knife.
Galaxy also separately opened a $5 million Bitcoin security fund, another sign that the sector is finally treating security as a hard requirement rather than a branding exercise.
The stakes are not theoretical. Immunefi reported roughly $110 million in crypto project losses to hacks in July and recorded 164 incidents through Aug. 3. It also projected that hacks exceeding $1 million could reach a record 114 in 2026. That forecast is only a projection, but it fits the trendline: attackers keep scaling, and the bill keeps landing somewhere real.
The coalition’s ask is narrow for a reason. It wants trusted access for vetted open-source security researchers, not unrestricted public use. That means the goal is to help people working on Bitcoin wallets, payment infrastructure, and other public software find vulnerabilities early, not to create a free-for-all where anyone can spin up a model and start probing for ways to steal funds.
The policy challenge is obvious. If access is too tight, defenders get stuck using second-tier tools while attackers experiment freely. If access is too loose, the same models meant to harden the system can be abused to break it. The sweet spot is controlled access with real verification, secure environments, and actual communication channels to the lab teams responsible for abuse response.
No leading AI laboratory had publicly announced a program responding to the coalition’s specific requests at the time of writing. That silence says plenty. AI companies are happy to talk about safety in the abstract. The harder part is giving defenders the kind of controlled access that is genuinely useful without turning it into an abuse pipeline with a compliance brochure stapled to it.
Key takeaways and questions
-
What are Bitcoin groups asking AI labs to do?
They want controlled access to frontier models for vetted open-source security researchers, plus compute, secure environments, and direct support from lab security teams. -
Why does Kimsuky matter here?
It shows attackers can use local AI tools outside cloud-based safety controls. That weakens the idea that platform guardrails alone are enough. -
Are AI-assisted vulnerability findings automatically trustworthy?
No. They still need human validation and reproduction. Big numbers are useful, but confirmed issues matter more than automated output. -
Is Bitcoin’s security problem just about AI?
No. AI is an amplifier, not the root cause. Bugs, weak entropy, operational mistakes, and plain old theft still do plenty of damage. -
Does post-quantum cryptography mean Bitcoin is in immediate danger?
No. It means serious players are planning for a future threat now, before quantum risk becomes a practical headache.
The bigger picture is hard to miss. Offensive AI is already in the wild, while defensive access is still rationed, delayed, or blocked by corporate caution. That is a lousy setup for any open system, especially one where public code, private keys, and human error all collide at high speed.
Bitcoin and the wider crypto stack were never going to stay secure on vibes alone. If the industry wants fewer hacks, fewer phishing campaigns, and fewer wallet disasters, it needs better tools in the hands of better defenders. Pretending the race does not exist would be stupid.
Related concerns are showing up well beyond Bitcoin, too. Australian regulators have warned that frontier AI could create larger risks for financial firms, while privacy advocates keep hammering the case for better financial privacy as a rare common ground among Americans. Meanwhile, quantum risk is no longer a sci-fi side quest. Researchers keep probing it, Coinbase has warned about possible long-term threats to Bitcoin wallet security, and Jameson Lopp has pushed an urgent five-year plan to prepare funds for a post-quantum world. For a technical deep-cut on the current debate, see Two More Waves Raise Suspected Coldcard-Linked Losses to, Quantum Hardware Breaks 15-Bit ECC Key, Renewing Bitcoin, Coinbase Warns Quantum Computing Could Threaten Bitcoin, Quantum Threat to Bitcoin: Jameson Lopp’s Urgent 5-Year, Australian banks warned frontier AI could create larger, , and Financial Privacy Is Rare Common Ground Among Americans.
