Anthropic’s restricted Claude Mythos preview is emerging as a defining case study in how frontier AI is being commercialized: not as a mass-market chatbot, but as a tightly controlled cybersecurity tool released only to trusted partners.
OpenAI said GPT-5.5 is now rolling out to paid ChatGPT and Codex users, while GPT-5.5 Pro is landing for higher-tier customers as the company continues expanding its GPT-5 line into professional workflows.
Anthropic’s Claude Mythos Preview is emerging as one of the clearest signs yet that generative AI is moving into a controlled-release era, where the most capable models may be withheld from public launch and shared only with selected security partners.
OpenAI is widening access to GPT-5.4-Cyber, a cyber-permissive version of its flagship reasoning model, as it scales its Trusted Access for Cyber program to more verified defenders. The change gives security teams a more capable model for legitimate reverse engineering and vulnerability analysis while keeping the rollout gated by identity and trust checks.
OpenAI has begun rolling out GPT-5.3 Instant Mini in ChatGPT, replacing GPT-5 Instant Mini as the fallback model for users who hit rate limits on GPT-5.3 Instant. The change is small on paper, but it matters for how ChatGPT behaves under load and after usage caps are reached.
