The Kraken parent company Payward disclosed that it had acceded to Project Glasswing of Anthropic and had started to use the Claude Mythos 5 clearance system for security loopholes. The results will be reviewed by the existing security team and repaired through established company processes.

First public encryption company

According to Payward, this was the first time that its security services had access to early access to such front-line cybersecurity models. According to public information, Payward was also the first encryption company to disclose its participation in the project.

Project Glasswing is a project by Anthropic for a cyber-defence scene that opens its high-capacity security model to audited institutions. The project was launched in April this year and expanded in June. Anthropic had previously stated that the participating agencies had identified thousands of high-risk or serious-level loopholes using relevant models.

The leaks will be handled by the internal team.

Payward states that Claude Mythos 5 will be used to scan potential gaps in the company ' s system and that the results will be sent to the internal security team for review before deciding how to repair them. The company will also inform the counterpart project maintainer if the issue concerns third-party open source software.

  • Scan object: In-house systems and related software
  • Disposition: advance repair after review by the security team
  • External collaboration: third-party open-source loopholes will provide feedback maintainers

According to the company, this approach is used not only to protect its own financial infrastructure but also to enhance the security of open-source software, which is widely used in the encryption industry.

Before the industry called for an open model.

This arrangement appears after the encryption industry has asked AI laboratories to open credible access. Just last week, more than 40 Bitcoins, including Kraken, and encryption companies sent letters to Anthropic, Openai and others calling for the most powerful models to be opened to qualified defensive security researchers.

The joint letter was organized by the Bitcoin Institute of Policy and involved Ark Invest, Coinbase, Block and BitGo. The letter stated that the developers responsible for protecting the open-source financial infrastructure needed to use Frontline AI to detect and repair loopholes earlier so that the attackers could not take advantage of them.

The CEO Arjun Sethi, co-chair of Payward, stated that traditional security defences were in a passive position for a long time, that only one loophole had to be found by the attackers, and that the defensive party had to detect all the problems as far in advance as possible. In his view, Front Line AI could review the code on the size of the machine, thereby helping the defensive to detect risks earlier.

Additional information:Mozilla disclosed in April this year that during the tests, Claude Mythos of Anthropic identified 271 loopholes in Firefox that show that these models are being used more for gap detection and safety tests.