Anthropic limits access to Mythos, its new cybersecurity AI model

Anthropic has introduced a cutting-edge cybersecurity AI model, Claude Mythos Preview, but access is strictly limited to an elite group of organizations. Notable companies such as Amazon, Apple, and Microsoft are among the chosen few who will utilize this advanced technology. In a move to ensure security, Anthropic is only granting access to thoroughly vetted entities, which also include tech giants like Broadcom, Cisco, and CrowdStrike. This announcement comes on the heels of troubling data leaks that have raised questions about Anthropic’s security measures. Just last month, sensitive information regarding the Mythos model surfaced online due to a data breach that revealed documents stored in a publicly accessible cache. Compounding the issue, another leak occurred recently, exposing internal source code for Anthropic's personal assistant, Claude Code. The company attributed these incidents to “human error,” which has amplified concerns regarding its data management practices. Despite the setbacks, Anthropic has been testing the Mythos model with partners for several weeks. While designed as a general-purpose AI, this is the first instance where the company has restricted the release of a model specifically due to its cybersecurity capabilities. The firm asserts that Mythos can uncover cyber vulnerabilities on an unprecedented scale, although it also possesses the potential to exploit those vulnerabilities, raising alarms about misuse by malicious actors. Dianne Na Penn, head of product management and research at Anthropic, emphasized the transformative potential of the technology, stating it could significantly enhance cybersecurity practices. However, she also warned of the risks involved if such powerful tools fall into the wrong hands. Selected companies will gain an advantage in identifying and securing vulnerabilities at a scale previously unattainable, underscoring the dual-edged nature of this advanced AI.