Key Developments

A significant data leak has exposed Anthropic’s development of Claude Mythos, a new AI model that reportedly represents “a step change” in AI performance. The leak occurred when nearly 3,000 internal documents became publicly accessible through an unsecured data store, revealing details about the “Capybara edition” of Mythos.

Meanwhile, OpenAI CEO Sam Altman announced internally that the company has completed pretraining of its new model codenamed “Spud,” with expectations of a “very strong model” ready within weeks. To support this launch, OpenAI is shutting down its video app Sora to free up computing resources.

Industry Context

The timing is significant as both companies are racing toward planned IPOs in 2026, creating intense pressure to demonstrate technological superiority. Claude Mythos reportedly shows “dramatic” improvements over previous models in programming tasks and reasoning, while excelling particularly at cybersecurity vulnerability detection.

The cybersecurity implications are substantial enough that shares of major security providers including CrowdStrike and Palo Alto Networks dropped over 5% following the news, suggesting market concerns about competitive disruption.

Practical Implications

For European organisations and developers, these developments signal a major shift in AI capabilities, particularly around cybersecurity. Anthropic’s focus on defensive applications suggests responsible deployment, but the leaked documents warn of “an impending wave of AI-driven exploits” that could outpace current defenses.

The models’ enhanced reasoning capabilities will likely impact software development workflows, with potential applications across industries from finance to healthcare. However, initial access appears limited to select enterprise clients, particularly for cybersecurity use cases.

Open Questions

Critical uncertainties remain around release timelines, pricing structures, and regulatory compliance with emerging EU AI legislation. The accidental leak raises questions about data security practices at leading AI companies, while the competitive timing suggests both firms may be rushing releases to market advantage.

For Irish and European organisations, the key question is how these capabilities will align with GDPR requirements and the EU AI Act’s risk assessment frameworks, particularly given the cybersecurity applications.

Source: The Information