A coalition of crypto firms and industry groups has urged frontier artificial intelligence labs to grant Bitcoin developers and other open-source defenders early, trusted access to their most capable models. The call comes in a letter published Monday by the Bitcoin Policy Institute (BPI), arguing that public access and “guardrails” on top-tier systems can leave key maintainers reliant on less capable alternatives.

In the letter, BPI and the signatories say many people responsible for maintaining critical digital infrastructure—including Bitcoin Core developers—may not have the ability to run high-end AI tools against complex codebases. That, they argue, can slow security research and reduce defenders’ ability to respond as threats evolve.

Key takeaways

• BPI says open-source financial infrastructure defenders often lack early access to frontier AI tools needed to keep pace with escalating cyber threats.
• The letter argues that guardrails on public frontier models can block qualified researchers from conducting effective security work.
• Signatories call for “standing trusted-access programs” for qualified maintainers of open-source financial infrastructure.
• BPI cites recent increases in crypto hacking activity and warns that AI-enabled attack techniques can increase risk for users.

Why the letter focuses on “trusted access”

The BPI letter frames frontier AI as a shift in how security research is performed. According to the letter, advanced models can scan large codebases more efficiently, flag potential weaknesses, and compress timelines for complex technical analysis—capabilities that can benefit both defenders and adversaries.

The core recommendation is practical: frontier AI labs should establish or expand “standing trusted-access programs” that allow qualified open-source financial infrastructure defenders to use high-performing models. Without such programs, the letter warns that defenders “may lack the tools needed to keep pace with evolving threats to the infrastructure they maintain.”

BPI also says it has received multiple independent reports from open-source maintainers describing sophisticated actors using advanced AI capabilities to support attacks. The implication is that defenders may be forced to work from a disadvantage if they cannot access the same level of AI capability under safe, controlled conditions.

Open-source infrastructure risk and why Bitcoin is central

The letter argues that open-source software underpins key parts of digital and financial systems. It singles out Bitcoin, stating that it alone secures more than $1 trillion in value. While the letter does not detail the measurement method, it uses that figure to emphasize the real-world stakes of maintaining and securing open-source infrastructure.

BPI further states that vulnerabilities in open-source infrastructure can endanger users’ life savings. That argument links the access request to a broader security policy question: how to balance model safety and guardrails with the need for qualified maintainers to conduct effective defense research.

Just as importantly, the letter suggests a mismatch between “publicly available” AI systems and the reality of defending production-grade infrastructure. If frontier tools are constrained such that certain security workflows are blocked, then—even for well-intentioned developers—defense capacity may not scale at the pace of attacker capabilities.

Crypto hacking surge underscores the pressure on defenders

The letter’s security pitch arrives alongside signs of mounting pressure across the broader crypto ecosystem. It points to DefiLlama data indicating that hacking activity across the industry surged in April 2026, when malicious actors reportedly stole more than $634 million from cryptocurrency platforms—described in the letter as the highest monthly total since the Bybit hack.

DefiLlama’s dashboard is cited in the letter for those figures, and it also notes that the Bybit incident contributed to total losses of roughly $1.4 billion in February 2025. While the letter does not attribute the April 2026 thefts to AI-enabled techniques, the juxtaposition is clear: as cyber incidents increase, defenders need better tooling and faster ways to assess and mitigate vulnerabilities.

For market participants who rely on infrastructure maintainers—exchanges, custody providers, wallet vendors, and protocol teams—the practical effect of slower vulnerability discovery can be significant. The difference between months and weeks can determine how quickly patches roll out, how quickly monitoring improves, and how much exposure a system carries before fixes reach production.

AI-enabled vulnerability discovery and the “vulnerability apocalypse” concern

The letter ties its access request to a broader trend in crypto security: AI-assisted vulnerability discovery is raising concerns across the industry. It references commentary from Mitchell Amador, CEO of bug bounty platform Immunefi, who described the current environment as a “vulnerability apocalypse,” in earlier coverage by Cointelegraph.

That earlier reporting cited the growing role of frontier models such as Claude Opus 4.8 and ChatGPT 5.5 in accelerating vulnerability research. The BPI letter uses that context to argue that advanced AI is increasingly part of the threat landscape—meaning defenders also need effective, timely access to advanced tools to conduct their own research and response.

Crucially, this is not framed as unrestricted model use. Instead, it centers on the idea that defenders should be able to work with frontier systems through trusted programs, designed to allow security research while reducing the risks associated with misuse.

Who signed the open letter

The open letter was co-signed by multiple crypto companies and organizations, including the African Bitcoin Institute, Anchorage Digital, BitGo, Bitwise, Blockstream, Bull Bitcoin, MARA, Kraken, Ledger, and Trezor, among others.

With this mix of infrastructure providers, custodians, security-oriented stakeholders, and Bitcoin-focused organizations, the letter reflects a common concern across the sector: that the security advantage could tilt toward attackers if AI capability is easier for adversaries to access than for open-source maintainers.

Going forward, the key question for readers is whether major AI labs respond by creating or expanding trusted-access programs that can be used by qualified open-source financial infrastructure defenders—and, if so, what eligibility and guardrail structures will look like in practice. The next signals to watch are concrete policy changes from frontier labs and measurable shifts in how quickly critical vulnerabilities are identified and patched as hacking activity remains elevated.

“If these processes can become faster and more efficient, U.K. businesses could unlock working capital sooner and make it easier to finance international trade,” Jacobsson said in an interview over LinkedIn.

The BOE named NOBO Finance, Dun & Bradstreet, a global provider of business decisioning data, analytics, and credit-rating services, and Polygon Labs, a software and blockchain company, as participants in its Digital Pound Lab.

The project will be the first time the Digital Pound Lab tests how public stablecoins and central-bank money work in a single payment flow alongside a portable credit identity for small businesses. The lab uses no real customers or money and does not signal any decision to issue a digital pound.

NOBO, a U.K.-based fintech building digital trade finance infrastructure that helps small and medium-sized enterprises (SMEs) become visible, verifiable, and bankable, was already involved in Phase 1. During the first phase, NOBO helped demonstrate conditional business-to-business escrow payments relevant to trade finance workflows.

A first workstream will build an SME “bankable profile.” NOBO, Dun & Bradstreet and Polygon plan to combine wallet transaction data, open-finance information and business intelligence to create a reusable credit assessment. Polygon will provide smart contracts intended to record the verified outcome and manage consent.

A group of crypto companies and industry organizations has called on frontier AI labs to give Bitcoin and other open-source financial infrastructure developers trusted access to their most capable models as AI-assisted cyber threats become more advanced.

The Bitcoin Policy Institute said in an open letter published Monday that developers responsible for securing open-source financial systems can be excluded from specialist cyber programs or restricted by safeguards built into publicly available frontier AI models.

Bitcoin developers seek access to stronger AI security tools

Under the proposal, AI companies would “establish or expand standing trusted-access programs for qualified defenders of open-source financial infrastructure,” allowing vetted security researchers and maintainers to use capabilities that may otherwise be restricted.

BPI argued that access has become more important as advanced AI systems gain the ability to examine large codebases, identify potential vulnerabilities and speed up difficult technical work. These capabilities can help legitimate researchers find flaws, but the institute said they can also be used by attackers looking for weaknesses.

For Bitcoin developers, the letter said the access gap can leave maintainers dependent on less capable open-weight models when frontier systems either refuse security-related requests or remain available only through programs that do not include open-source financial infrastructure.

The institute pointed to the amount of capital dependent on such software, noting that Bitcoin alone secures more than $1 trillion in value. A serious vulnerability in financial infrastructure can therefore put users’ savings at risk, the letter said.

Several major crypto companies joined the request, including Anchorage Digital, BitGo, Bitwise, Blockstream, Bull Bitcoin, MARA, Kraken, Ledger and Trezor. The African Bitcoin Institute was also among the organizations that signed the letter.

Bitcoin Core’s recent security work has provided examples of the type of vulnerabilities maintainers must identify before they can affect users. In June, crypto.news previously reported that Bitcoin Core 31.1rc1 fixed a privacy problem involving PrivateBroadcast that could expose a user’s IP address under certain network conditions. The release candidate also contained changes covering wallet accuracy, networking, blockchain validation, and MuSig2 security.

A month earlier, Bitcoin Core developers disclosed a high-severity bug tracked as CVE-2024-52911 that could allow miners to remotely crash some nodes. The vulnerability affected versions after 0.14.0 and before 29.0, although triggering it required miners to produce costly proof-of-work blocks. Security researcher Cory Fields privately reported the flaw in 2024 before Bitcoin Core 29.0 shipped with the fix in April 2025.

Frontier AI could strengthen open-source defenders

BPI described frontier AI as a technology that is changing the economics of both cybersecurity research and cyber operations because increasingly capable models can perform tasks that previously required substantial amounts of specialist human work.

Advanced models could eventually become one of the “most powerful defensive technologies ever developed,” the institute said, particularly when researchers use them to inspect software and identify weaknesses before attackers exploit them.

“Without dedicated access programs, defenders may lack the tools needed to keep pace with evolving threats to the infrastructure they maintain,” the letter said.

BPI also said it had received multiple independent reports from open-source maintainers about sophisticated actors using advanced AI capabilities to sustain attacks. Some of the activity potentially involved foreign adversaries, according to the institute.

Rather than asking AI developers to remove security controls from their models for all users, the signatories are seeking standing programs through which qualified defenders could receive additional access after being vetted.

The proposal would put open-source financial developers closer to researchers and organizations already permitted to use advanced cyber capabilities under controlled programs, while retaining restrictions intended to prevent unrestricted access by malicious users.

The potential defensive value of AI has already been tested elsewhere in crypto. In July, an Ethereum Foundation study found that coordinated AI agents could uncover genuine vulnerabilities in software used by Ethereum, including a libp2p flaw later disclosed as CVE-2026-34219. The Foundation’s Protocol Security team said the harder part was determining which AI-generated reports represented real vulnerabilities rather than convincing false positives.

Human validation and reproducible proof therefore remained necessary even when AI systems successfully identified security problems, according to the Foundation.

AI-assisted attacks raise pressure on crypto security

The request comes after crypto platforms suffered another heavy period of security losses in 2026, with attackers increasingly able to combine software vulnerabilities, compromised credentials, social engineering and other attack methods.

DefiLlama data cited in June showed more than $634 million was stolen from cryptocurrency platforms during April, the industry’s highest monthly loss since the Bybit hack, which contributed to roughly $1.4 billion in losses in February 2025.

Earlier figures covered in April showed more than $606 million had already been stolen across 12 incidents during the first 18 days of the month. The amount was roughly 3.7 times the $165.5 million lost throughout the first quarter of 2026.

Two attacks accounted for most of April’s losses at that stage. Drift Protocol lost about $285 million, while an exploit involving KelpDAO resulted in losses of roughly $292 million, with the two incidents accounting for about 95% of the reported total during the first 18 days.

The attack pattern has also moved outside isolated smart-contract bugs. DefiLlama had recorded more than $17 billion in losses across 518 crypto hacking incidents over the previous decade by April, with private-key leaks, phishing and credential theft accounting for an increasing part of the damage alongside protocol exploits.

At the same time, security companies have warned that AI can lower the amount of time and technical work required to search for exploitable weaknesses. CertiK said in April that AI-assisted phishing, deepfakes and automated exploit tools were making attacks faster and harder to detect, while cross-chain infrastructure and social engineering remained important attack routes.

Advanced AI models have changed vulnerability discovery

Concerns over access have intensified as frontier models demonstrate stronger vulnerability-discovery capabilities, giving security researchers tools that can inspect software at a scale that was previously difficult to achieve.

Mitchell Amador, CEO of bug bounty platform Immunefi, described the proliferation of systems including Claude Opus 4.8 and ChatGPT 5.5 as contributing to a “vulnerability apocalypse” for crypto security, arguing that advanced models had changed the balance between attackers and defenders.

Amador said the next three to four years could be critical for crypto cybersecurity while defensive teams work to use the same AI capabilities to produce more secure codebases. Increased use of crowdsourced security systems could shorten that period to less than two years, he added.

AI-based security tools were already part of the industry’s defensive infrastructure before the latest concerns. In an October 2025 interview with Immunefi, Amador said automated vulnerability detection should operate alongside audits, bug bounties, monitoring systems and transaction firewalls rather than replace them. He said at the time that fewer than 10% of projects used AI vulnerability tools.

Ethereum co-founder Vitalik Buterin made a similar case for defensive uses in May, arguing that AI-assisted formal verification could eventually allow developers to combine highly optimized software with machine-checked proofs of correctness. He cited potential applications across Ethereum’s consensus systems, zero-knowledge technology, and quantum-resistant cryptography while cautioning that formal verification could not eliminate every source of software risk.

CoreWeave (CRWV) shares surged 16% in pre-market trading on Wednesday after the company delivered stronger growth and raised its full-year outlook, lifting the broader neocloud market.

The AI infrastructure provider said second-quarter revenue more than doubled to $2.58 billion, narrowly beating expectations, while its net loss of $626 million was smaller than analysts projected. The company forecast third-quarter sales of $3.45 billion to $3.6 billion and increased its 2026 revenue guidance to between $12.4 billion and $13.2 billion.

IREN (IREN) and Cipher Digital (CIFR), both gained 5% as investors piled into companies positioned to benefit from accelerating demand for AI computing.

CoreWeave ended the quarter with $104 billion in contracted business and added more than $25 billion of customer commitments after the period closed. Management also said recently signed deals carried margins five to 10 percentage points above recent levels, reflecting scarce capacity and favorable pricing.

The results reinforce a widening market divide. AI infrastructure continues to attract capital, customers and premium valuations, while bitcoin and the broader crypto trade struggle to match that momentum as bitcoin trades below $64,000.

Fidelity is preparing to add staking and quarterly cash payouts to its Fidelity Ethereum Fund (FETH), one of the largest spot ether ETFs in the U.S.

FETH, with $898 million in net assets, could stake as much as 100% of its ether under normal conditions, though Fidelity set no minimum, according to an amended registration statement. The fund would keep some ETH available for redemptions, expenses and other liquidity needs.

The shift follows an IRS safe harbor bulletin issued in November 2025 that lets qualifying crypto trusts stake assets without losing their grantor-trust tax status. Fidelity would join Grayscale and 21Shares in adding staking to existing ether funds. BlackRock took a different route by introducing a separate staking product.

Fidelity would retain 85% of gross staking rewards, while the remaining 15% would go to the fund sponsor, custodians and node operators. Blockdaemon, Figment and Galaxy are named as the trust’s node operators.

Net staking rewards would first cover fund expenses and would then be used for quarterly cash distributions. Funds must distribute net staking rewards at least quarterly, according to the IRS rules.

The fund may also sell some ETH to raise cash for payouts, Fidelity said.