OpenAI repositioned Codex as a general-purpose work assistant, moving beyond its coding origins to target knowledge workers across industries.

The company launched "Codex for Work" with 42% faster computer use automation, responsive browser capabilities, and new workflow commands including /chronicle and /goal. The updated agent now integrates with Microsoft Office, Google Workspace, and Salesforce, featuring an in-app file editor for Office documents.

"Codex is for everyone, for any task done with a computer," OpenAI's Greg Brockman said on X. The interface dynamically adapts based on the task rather than offering manual toggles, according to product manager Alex Jambrosino.

The timing coincides with new UK AI Security Institute evaluations showing GPT-5.5 achieved rough parity with Claude Mythos Preview on multi-step cyber-attack simulations. GPT-5.5 recorded a 71.4% average pass rate versus 68.6% for Mythos, marking the second model to complete end-to-end cyber scenarios.

Anthropic countered with Claude Security, a code review tool addressing rising security vulnerabilities in AI-generated code. The launch comes amid what security researchers describe as a "meta mythos" around the Claude Mythos model's capabilities.

More significantly, Claude added support for creative applications including Blender, Autodesk, Adobe Creative Cloud, Ableton, Splice, and Canva. The integration allows the AI assistant to control creative workflows directly within professional design and audio production software.

The creative tools expansion represents Claude's bid to differentiate from OpenAI's enterprise-focused positioning. While Codex targets business productivity workflows, Claude is positioning itself as the AI assistant for creative professionals.

Both companies are racing to expand their agents beyond narrow use cases. The UK cyber evaluations suggest the capability gap between frontier models is narrowing, potentially accelerating the competition for specialized applications.

OpenAI also shipped Advanced Account Security for ChatGPT, adding phishing-resistant authentication and hardened account recovery as AI models demonstrate increasing offensive cyber capabilities.