Anthropic's Mythos is evolving faster than expected, reports AI safety agency

The update comes just one month after Mythos' initial release. The institute's findings indicate that the model's capabilities are evolving more rapidly than previously anticipated.

The report highlights that the newer Mythos iteration achieved superior performance on benchmark tests compared to both its predecessor and GPT-5.5, which is OpenAI's most advanced publicly known model. The exact benchmarks used and the magnitude of the performance difference were not detailed in the excerpt. The UK AI Security Institute is tasked with evaluating frontier AI models for potential risks.

This development underscores the pace of progress in the AI field, where models can see significant improvements within weeks. It also positions Anthropic's Mythos as a leading model, at least temporarily, ahead of OpenAI's offering. The UK AI Security Institute continues to monitor and test such models as part of its mandate, though the implications of this rapid evolution for safety and regulation remain to be fully assessed.