Anthropic and OpenAI Ignite 'Cyber AI' Arms Race with Gated Frontier Models
In a pivotal week for AI safety and competition, Anthropic released Claude Opus 4.7 for general use while keeping its most powerful model, 'Claude Mythos,' in a restricted preview due to high cybersecurity risks. OpenAI countered with the release of GPT-5.4-Cyber, a specialized flagship variant tuned for defensive security. This shift marks a new industry standard where 'frontier-class' capabilities are increasingly gated behind vetted access programs like Anthropic’s Project Glasswing and OpenAI’s Trusted Access for Cyber.
OpenAI launched GPT-Rosalind, its first specialized model for life sciences, aimed at accelerating drug discovery and genomics.
Meta and Broadcom announced a massive partnership to co-develop multiple generations of custom MTIA AI silicon, with power commitments exceeding 1GW.
Google DeepMind released Gemini Robotics-ER-1.6, a Vision-Language-Action model designed to help robots reason through complex physical tasks.
Meta is reportedly testing an internal 'AI Zuck' clone—a personal agent trained on Mark Zuckerberg's mannerisms to attend meetings and advise staff.
OpenAI updated its Agents SDK to include native sandbox execution, allowing AI agents to safely edit code and run system commands in controlled environments.
Did you know?
Boston Dynamics’ Spot robot, now powered by Google DeepMind’s Gemini Robotics-ER 1.6, can read analog gauges and thermometers with 98% accuracy during autonomous industrial inspections.
Quiz
Which restricted Anthropic model was deemed 'too dangerous for public release' this month due to its advanced cybersecurity capabilities?