Executive Overview
In response to this shifting threat landscape, a coalition of open-source developers and security researchers formed the Bitcoin Red Team. Operating as an emergency defense collective, this volunteer force is racing against malicious actors to locate, classify, and patch vulnerabilities across the broader Bitcoin software ecosystem before they can be exploited at scale.
While the core Bitcoin protocol layer remains resilient, the software peripheral to the blockchain—including wallet applications, layer-2 protocols, exchange interfaces, and custodial tools—faces an unprecedented wave of automated threat activity. To combat this threat, the Bitcoin Red Team has deployed advanced AI models to proactively scan open-source repositories. Paradoxically, the team relies heavily on open-weights Chinese AI models, such as Moonshot AI’s Kimi K3, due to aggressive safety guardrails embedded in Western AI systems that unintentionally obstruct legitimate defensive cybersecurity research.
Detailed Chronology
[Coldcard Wallet Incident]
│
▼
[Rob Hamilton / Security Audits Initiated]
│
▼
[Emergence of Advanced AI Models (Kimi K3)]
│
▼
[Formation of Bitcoin Red Team (~20–25 Volunteers)]
│
▼
[Full Ecosystem Proactive Code Sweeps & Dev Feedback]1. The Catalyst: High-Profile Exploits and Asset Migrations
The initial momentum for a dedicated defensive task force accelerated following a security exploit involving Coldcard air-gapped hardware wallets. The incident exposed critical operational vulnerabilities in peripheral software, forcing defensive responses across the industry that included moving over $1.5 billion in Bitcoin assets to secure storage as a precautionary measure. The breach highlighted a structural vulnerability: even if a core protocol is cryptographically sound, the software applications bridging users to the network remain vulnerable to attack.
2. The AI Shift: Arrival of Kimi K3
Shortly after the Coldcard incident, the public release of advanced reasoning models—most notably Moonshot AI’s Kimi K3 model—altered the operational calculus for both cybersecurity defenders and threat actors. Kimi K3 demonstrated advanced capabilities in code analysis, multi-step logical deduction, and sandbox navigation. This marked a turning point where automated systems could scan thousands of lines of open-source repository code to identify logic flaws and memory corruption bugs in minutes—tasks that previously required days of manual code auditing.
3. Formation of the Red Team
Recognizing that threat actors were already leveraging these tools, Rob Hamilton, CEO of Bitcoin insurance firm AnchorWatch, began auditing peripheral Bitcoin infrastructure projects. Simultaneously, pseudonymous Bitcoin developer Calle—a maintainer of the open-source Cashu protocol—and a core group of privacy and protocol researchers mobilized an emergency response unit. The group quickly grew to include between 20 and 25 volunteer developers, establishing a war room to organize proactive scans across the open-source Bitcoin landscape.
4. Proactive System Sweeps and Escalation
By mid-2026, the team escalated its operations from reactive patch verification to proactive ecosystem-wide sweeps. Utilizing automated script pipelines powered by high-capacity LLMs, the group systematically audited open-source code repositories across wallets, Cashu implementations, Lightning Network integrations, and node-management software. Spending over $20,000 in API compute costs—fully underwritten through private industry backing—the group established a direct feedback loop with affected developers to quietly disclose vulnerabilities and refine severity classification models before public exposure could occur.
Supporting Context & Technical Metrics
The Erosion of "Security Through Obscurity"
For decades, open-source software security operated on the assumption that discovering complex vulnerabilities required significant domain expertise, creating a natural friction against widespread exploitation. The emergence of reasoning AI models has effectively eliminated this information asymmetry.
As AI lowers technical barriers, an attacker with minimal formal training in C++, Rust, or Python can feed public code bases into an unaligned model with instructions to locate exploitable memory leaks, logic bugs, or improper inputs. Once identified, the AI can assist in constructing functional exploit payloads. Consequently, software maintainers can no longer rely on security through obscurity or assume that obscure bugs will remain unnoticed in public code bases.
TRADITIONAL THREAT MODEL:
[High Skill Requirement] ──► [Manual Code Audit] ──► [Slow Exploit Discovery]
CURRENT AI-DRIVEN THREAT MODEL:
[Low Skill Requirement] ──► [Automated AI Sweeps] ──► [Rapid Exploit Generation]The US vs. China AI Guardrail Paradox
One of the key operational insights revealed by the Bitcoin Red Team is their reliance on Chinese AI models over leading American alternatives. Frontier models developed by US firms (such as Anthropic and OpenAI) incorporate strict safety protocols designed to prevent the generation of malicious code or vulnerability analysis.
However, these guardrails frequently trigger false positives during legitimate, white-hat security research:
- Refusal of Vulnerability Identification: US models often refuse to review code blocks if the prompt asks to locate "exploitable vulnerabilities" or "security flaws."
- Refusal of Remediation: Researchers report instances where US models refused to assist in writing patches for already identified vulnerabilities, misinterpreting the patch logic as a method to study the exploit vector itself.
- Shift to Unrestricted Alternatives: Chinese open-source and open-weights models, such as Moonshot AI’s Kimi K3, DeepSeek, and MiniMax, lack these restrictive filters, allowing white-hat security researchers to perform deep automated vulnerability assessments without model refusal.
This dynamic exists alongside heightened geopolitical friction surrounding artificial intelligence capabilities. Earlier in the year, Anthropic publicly accused several Chinese AI laboratories—including DeepSeek, Moonshot AI, and MiniMax—of operating roughly 24,000 fraudulent accounts to extract over 16 million exchanges from Claude models via model distillation. Shortly thereafter, the Trump administration issued warnings asserting that foreign entities were engaging in the industrial-scale extraction of American AI intellectual property.
Despite these geopolitical dynamics, white-hat security groups like the Bitcoin Red Team frequently choose Chinese models due to operational utility, as overly restrictive Western guardrails can inadvertently hinder defensive cybersecurity efforts.
+------------------------------------+------------------------------------+
| US Frontier AI Models | Chinese AI Models (Kimi K3, etc.) |
+------------------------------------+------------------------------------+
| • High reasoning capabilities | • Advanced technical capabilities |
| • Restrictive safety guardrails | • Minimal cybersecurity filters |
| • Rejects white-hat code scans | • Enables automated vulnerability |
| • Hinders patch development | auditing and white-hat scans |
+------------------------------------+------------------------------------+Operational Scope and Financial Metrics
- Team Composition: ~20 to 25 volunteer developers and security researchers.
- Key Contributors: Pseudonymous developers Calle (Cashu), Stu, Talip, and
thesimplekid; alongside Bitcoin developers Ben Carmen, Daniela Brozzoni, and James O’Beirne, and Vinteum Bitcoin R&D Center board member Bruno Garcia. - Operational Infrastructure Expenditure: Over $20,000 incurred in API compute costs, fully funded by internal industry backers without public donations.
- Ecosystem Coverage: Comprehensive sweeps covering the majority of active, open-source Bitcoin application software, including wallets, layer-2 mints, and transaction protocols.
Official Statements
Calle on the Critical Race Against Threat Actors
Calle, pseudonymous developer and Cashu protocol maintainer, emphasized the time-sensitive nature of the Red Team’s defensive campaign when speaking with reporters:
"At this point, it is a question about time. The reason why the Bitcoin Red Team exists right now is because we need to get ahead of the attackers as fast as possible."
Addressing the operational boundary between core blockchain infrastructure and edge software, Calle clarified that the protocol itself remains sound:
"Although Bitcoin itself is secure, the software that we’re using to transact with Bitcoin may not be, and that is what most people interface with anyway."
Calle on the Technological Shift and AI Asymmetry
Calle noted that the structural dynamics of cybersecurity changed dramatically following recent advancements in reasoning models:
"I think the arrival of Kimi K3 has also caused a lot of chaos in the cybersecurity realm because it gave attackers as well as defenders unprecedented power."
On the practical necessity of utilizing unguardrailed Chinese models over Western counterparts, he noted:
"It’s not even close. Although U.S.-based frontier models are still arguably more intelligent than any other models out there in the world, they all come with heavy guardrailing, which limits their use, especially in the cybersecurity realm."
Regarding the permanent loss of security through obscurity in open-source repositories:
"I think that there are no secrets anymore in software. There is no information asymmetry that was previously being used to kind of create security theater or security through obscurity. Those times are over. Simple exploits can now be completed end to end by someone who doesn’t know how to do it without AI. So AI gave people a form of power that has completely changed the playing field."
Rob Hamilton on Operational Mobilization
In an operational update detailing the funding and progress of the initiative, AnchorWatch CEO Rob Hamilton stated via social media:
"We have been working around the clock, with ~$20,000 of spend up to this point across different services. Funding is secured, I appreciate all the gestures for donations but it is not necessary. The bill is taken care of."
Future Outlook
Cryptocurrency as the Vanguard for Global Cyber Warfare
The security dynamics currently affecting the Bitcoin software ecosystem serve as a testing ground for broader enterprise software security. Cryptographic financial networks represent an immediate, liquid target for attackers. Unlike traditional software ecosystems where exploiting a vulnerability requires multi-stage network intrusions or data exfiltration, compromising a cryptocurrency application can allow immediate, irreversible financial extraction.
As a result, cryptocurrency developers are experiencing the first wave of AI-driven security threats. As AI tools continue to improve, legacy industries—including traditional banking, healthcare, legacy enterprise software, and critical infrastructure—will likely face similar automated vulnerability discovery challenges.
Strategic Imperatives for Open-Source Maintenance
To adapt to this environment, open-source software management will likely require structural changes:
- Continuous Automated Red-Teaming: One-time code audits will need to be replaced by persistent, automated AI security pipelines integrated directly into continuous integration and deployment (CI/CD) workflows.
- Re-evaluating AI Safety Policy: Western AI providers may face pressure to re-evaluate how safety filters are implemented, creating secure environments where verified white-hat researchers can utilize unredacted code-analysis models without triggering false-positive safety locks.
- Dedicated Open-Source Security Funding: Defensive security initiatives will require dedicated, long-term institutional funding to subsidize the compute costs associated with running continuous, ecosystem-wide AI vulnerability sweeps.
The emergence of the Bitcoin Red Team marks a significant shift in defensive cybersecurity. As AI continues to democratize technical capabilities, maintaining system integrity will increasingly depend on white-hat developers’ ability to automate defense faster than adversaries can automate exploitation.
