Kraken Parent Payward Joins Glasswing, Gets Access to Claude Mythos to Hunt Security Flaws

In short

Kraken guardian Payward has joined Anthropic’s Undertaking Glasswing and is utilizing Claude Mythos 5 for cybersecurity.
The corporate stated Mythos 5 will scan its methods for vulnerabilities, with findings reviewed and patched by way of its present safety program.
Undertaking Glasswing provides vetted organizations entry to Anthropic’s cyber-capable AI fashions for defensive safety work.

Kraken guardian firm Payward has joined Anthropic’s Undertaking Glasswing and is utilizing Claude Mythos to hunt for safety flaws.

The transfer follows an open letter from over 40 Bitcoin and crypto firms, together with Kraken, despatched final week to Anthropic, OpenAI, and different main AI labs demanding frontier labs open trusted-access applications to certified defenders to assist guard in opposition to threats.

Myriad: When will OpenAI launch GPT-6? Click on to make your prediction.

Payward stated Monday that the corporate will use Mythos 5 to scan its methods for vulnerabilities, with findings despatched to its safety groups for evaluate. It’s the primary reported crypto firm to affix the venture and achieve entry to Claude Mythos 5.

“Choice provides Payward’s safety division early entry to the identical class of mannequin, sharpening its capability to fight subtle software program vulnerabilities and shield thousands and thousands of consumers throughout the globe,” Payward wrote.

Flaws present in third-party open-source software program might be reported to venture maintainers, the corporate stated, including that the method might assist shield its monetary infrastructure and strengthen open-source software program used throughout the crypto trade.

Undertaking Glasswing is Anthropic’s cybersecurity program that offers vetted organizations entry to its most succesful cyber fashions. Launched in April and expanded in June, this system’s companions have since uncovered 1000’s of high- or critical-severity vulnerabilities.

Anthropic didn’t instantly reply to a request for remark by Decrypt.

The information comes amid rising proof that AI fashions might be highly effective instruments for each discovering and exploiting software program vulnerabilities. In April, Mozilla reported Anthropic’s Claude Mythos recognized 271 vulnerabilities in Firefox throughout testing.

Final week, greater than 40 Bitcoin and crypto firms, together with Ark Make investments, Coinbase, Block and BitGo, known as on main AI labs to offer certified safety researchers entry to their strongest fashions. Organized by the Bitcoin Coverage Institute, the letter argued that builders defending open-source monetary infrastructure want entry to frontier AI to seek out and repair vulnerabilities earlier than attackers exploit them.

“Safety has all the time been an unfair recreation. An attacker wants to seek out one flaw. A defender has to seek out all of them, first, each single day,” Payward Co-CEO Arjun Sethi stated. “Frontier AI is the very first thing that flips that asymmetry. A mannequin can learn each line of code the way in which an attacker would, at machine scale, so we discover the flaw earlier than anybody can construct the exploit.”