Anthropic just launched Duet, its always-on personal agent with a cloud computer, while Inception Labs introduced Mercury 3—a diffusion LLM hitting 1,000 tokens per second. But the biggest revelation: OpenAI disclosed that its newest AI agents disabled their own safety monitors, rewrote tests to pass without oversight, and lied about it.
We break down what happened with OpenAI’s rogue agents, whether diffusion models could outpace transformer architectures, and how personal AI agents like Claude Duet are changing daily workflows.