Overview
- Payward, the parent company of Kraken, has become a participant in Anthropic’s Project Glasswing and is utilizing Claude Mythos 5 for cybersecurity purposes.
- This initiative will allow the company to identify system vulnerabilities, which will then be addressed through its existing security protocols.
- Project Glasswing provides selected organizations with access to Anthropic’s advanced AI models aimed at enhancing defensive cybersecurity measures.
Payward, the parent organization of Kraken, has joined Anthropic's Project Glasswing, employing Claude Mythos to identify security weaknesses.
This decision follows an open letter sent last week by over 40 Bitcoin and cryptocurrency firms, including Kraken, to Anthropic, OpenAI, and other prominent AI laboratories. The letter urged these leading AI entities to establish trusted-access programs for qualified defenders to bolster security against potential threats.
Myriad: When will OpenAI release GPT-6? Click to make your prediction.On Monday, Payward announced that it will leverage Mythos 5 to conduct scans for vulnerabilities within its systems, with the results being forwarded to its security teams for further evaluation. Payward is the first cryptocurrency company to join this initiative and gain access to Claude Mythos 5.
“Being selected grants Payward’s security team early access to this class of model, enhancing its capacity to tackle complex software vulnerabilities and safeguard millions of customers worldwide,” stated Payward.
Any vulnerabilities detected in third-party open-source software will be reported to the relevant project maintainers, according to the company. This proactive strategy aims to protect its financial infrastructure and fortify the open-source software utilized across the cryptocurrency sector.
Project Glasswing is Anthropic's cybersecurity initiative that provides vetted organizations with access to its most capable cyber models. Launched in April and expanded in June, partners of the program have already identified thousands of vulnerabilities with high or critical severity.
Anthropic has not yet responded to a request for comments from Decrypt.
This development comes amidst increasing evidence that AI models can serve as effective tools for both detecting and exploiting software vulnerabilities. In April, Mozilla announced that Anthropic’s Claude Mythos had identified 271 vulnerabilities in Firefox during its assessments.
Last week, a coalition of over 40 Bitcoin and cryptocurrency companies, including Ark Invest, Coinbase, Block, and BitGo, appealed to major AI laboratories to grant qualified security researchers access to their most advanced models. This initiative, organized by the Bitcoin Policy Institute, contended that developers safeguarding open-source financial infrastructures require access to cutting-edge AI to preemptively identify and rectify vulnerabilities before they can be exploited by malicious actors.
“Security has always been an uneven playing field. An attacker only needs to discover one flaw, while a defender must uncover all of them consistently,” remarked Payward Co-CEO Arjun Sethi. “Frontier AI shifts this imbalance. A model can analyze every line of code as an attacker would, at a machine scale, enabling us to detect flaws before anyone can create an exploit.”