Anthropic's Mythos AI Model Launches Amid Security Leaks and Legal Tensions

Anthropic, an AI startup, has introduced a preview of its advanced AI model called Mythos, touted as one of its “most powerful” models. This introduction is part of Project Glasswing, a cybersecurity initiative involving 12 partner organizations including major tech companies like Amazon, Apple, and Microsoft.

• Mythos was not specifically designed for cybersecurity but will be applied to identify code vulnerabilities in software, and it’s noted for identifying thousands of decade-old zero-day vulnerabilities.
• The release is exclusive, with 40 organizations gaining preview access beyond the primary partners. A broader release is not anticipated currently.
• The introduction of Mythos occurs against a backdrop of legal disputes with the Trump administration, which classifies Anthropic as a supply-chain risk over potential AI usage controversies.
• The model’s inception came to light through a data leak due to human error, with a draft revealing the model was initially called “Capybara,” described as stronger than past models.
• The disclosure also exposed a vulnerability: the potential misuse of the model for malicious purposes in cybersecurity.
• Recent errors also led to exposure of Anthropic’s source codes, further complicating security efforts.