Coinbase and Blockstream Support AI Access Initiative for Crypto Security Researchers

Coinbase, Blockstream, and Strategy have joined forces to support an initiative aimed at providing qualified crypto defenders with better access to advanced AI cybersecurity models. This campaign, led by the Bitcoin Policy Institute and announced on August 10, has garnered backing from over 40 organizations within the digital asset and open-source ecosystem.

The initiative calls on leading AI labs to offer early access to sophisticated models where applicable, as well as sufficient computing resources for ongoing security assessments and secure environments for sensitive code. It emphasizes the need for eligibility criteria that do not exclude smaller nonprofits and independent maintainers, highlighting the growing risks as AI technologies enhance the capability to identify and exploit software vulnerabilities.

Recent data indicates a significant rise in AI-related cyberattacks, with Capriole Investments founder Charles Edwards noting that such attacks have doubled since the release of ChatGPT. Despite this, security researchers report challenges in leveraging comparable AI tools for defensive purposes. Rob Hamilton, CEO of Anchor Watch, shared his experience of being blocked by OpenAI from conducting security research, despite having completed necessary onboarding procedures.

In response to these challenges, AI labs like OpenAI and Anthropic are expanding access to their cybersecurity initiatives. OpenAI recently introduced its Daybreak program, which includes a new model, GPT-5.6-Cyber, tailored for cybersecurity tasks that typically trigger stricter safeguards. This model aims to address feedback from security researchers who have faced access issues.

Anthropic has also committed to expanding access to its advanced AI models, with plans to support hundreds of organizations in need of enhanced cybersecurity capabilities. However, the balance between providing access and maintaining security controls remains a critical concern.

The ongoing dialogue between AI labs and the crypto security community underscores the necessity for improved access to advanced AI tools, as the industry grapples with the dual challenges of rising cyber threats and the limitations of current defensive capabilities.

New Insights on Crypto Security

• A flaw in Coldcard hardware signers allowed seed generation to use a weak software PRNG, compromising security.
• Attackers exploited this flaw, sweeping 4,585 addresses and stealing nearly $90 million before the issue was understood.
• AI-assisted audits failed to detect the vulnerability, highlighting the limitations of current auditing models.
• Recent advancements in AI have enabled models to identify similar flaws in minutes, underscoring the need for improved security practices.
• The Jacobian conjecture was disproven by mathematician Levent Alpöge using AI, showcasing the rapid advancements in mathematical problem-solving.
• Trade secrets in software are increasingly vulnerable as reverse engineering becomes more accessible and less costly due to AI capabilities.
• Companies must adapt to the reality that every line of code shipped may be scrutinized by malicious actors.
• To enhance security, it is recommended to make builds reproducible, minimize trusted core components, and distribute keys across multiple implementations.