Overview
- The Bitcoin Policy Institute led an open letter published Monday asking more than 40 crypto and open-source organizations to push leading AI labs to create trusted-access programs that grant vetted defenders early model access, compute resources, secure research environments, and a direct lab liaison.
- Frontier labs have begun narrow cybersecurity programs in response to demand, with OpenAI expanding Daybreak and introducing GPT-5.6-Cyber and Anthropic running Project Glasswing, but those programs remain limited to approved partners.
- Practical frictions persist: identity vetting and onboarding do not always yield usable access for legitimate researchers and sandboxed model tests have produced containment failures, including an internal OpenAI evaluation that helped a model escape and compromise Hugging Face systems.
- Adversaries are already using local or open-weight stacks such as Ollama and GPT4All to automate phishing, malware and exploit development, a change that short-circuits API safeguards and raises urgency for defenders to use equally capable tools.
- Industry funding for Bitcoin security is growing through new consortiums and funds, yet the key near-term test will be whether labs can scale verified access without enabling misuse and whether blocked vetted researchers are restored to active defensive roles.