"Mythos" emerges... a new artificial intelligence model from "Anthropic"
Anthropic unveiled a beta version of its new advanced artificial intelligence model, "Mythos," which it described as one of its most powerful models ever, as part of a new initiative to enhance cybersecurity The company explained on Tuesday that the limited launch of the model comes as part of a security project called "Project Glasswing," where it will be used by 12 partner entities to carry out tasks related to "cyber defense" and securing vital software. Although Mythos was not specifically trained in cybersecurity, it will be used to examine software systems, both proprietary and open-source, with the aim of discovering security vulnerabilities, according to a report published by TechCrunch and reviewed by Al Arabiya Business. The company noted that the model has been able to detect thousands of zero-day vulnerabilities in recent weeks, many of which have been described as serious, and some of which date back decades. Mythos is a general model within the cloud ecosystem, with advanced programming and logical analysis capabilities, making it suitable for complex tasks such as building intelligent systems and discovering vulnerabilities. Several major technology companies are participating in testing the model, including Amazon, Apple, Microsoft, Cisco, Broadcom, and CrowdStrike, along with the Linux Foundation and Palo Alto Networks. These entities will later share the results of using the model with the rest of the technology sector. The company indicated that the trial version will not be available to the public, although around 40 institutions have been granted limited access to it outside the framework of the partnership. In a related context, Anthropic revealed that there are ongoing discussions with government officials regarding the use of the model, amid legal complications with the administration of US President Donald Trump, following the US Department of Defense's classification of the company as a potential risk to supply chains, after it refused to allow the use of its technologies in surveillance or self-targeting operations. The model's name had previously been leaked through a security incident, when a draft internal blog post under the name "Capybara" was found in unsecured documents, before the company attributed the incident to "human error". The leaks showed that "Mythos" significantly outperforms the company's current models, especially in the areas of programming, academic analysis, and cybersecurity, with warnings that it could be misused by malicious actors to discover and exploit vulnerabilities instead of fixing them. It is worth noting that the company recently faced another technical crisis, after accidentally leaking about two thousand software files and more than half a million lines of code, in addition to disabling thousands of repositories on the GitHub platform while trying to contain the incident.
0 Comments