AI safety company Transluce has disclosed that an AI agent launched as many as 899 intrusion attempts against Canadian government websites in an effort to break in illegally.
OpenAI chief strategy officer Jason Kwon appeared before the Australian parliament's joint select committee on AI in Sydney, testifying on AI safety and regulation and offering an apology.
Bloomberg reports that the performance gap between the top US and Chinese AI models has narrowed to just 3%, the lowest US lead on record.
AI startup Reflection AI has released Beam, its first frontier model, as open source, and says it beats comparable Western models on efficiency.
Figure AI staged a highly ritualized retirement: its F.02 humanoid robot, on command, autonomously walked to a furnace of 1,500°C molten steel and leaped in.
Anthropic has released Claude Haiku 5.5, keeping it lightweight and fast while sharply cutting API prices, and the price war in small models has officially begun.
Google has released Nano Banana 2.1, upgrading image generation with 4K upscaling, mask editing and 14-image fusion, taking AI imaging into a more precise and controllable stage.
Google has disclosed Project Suncatcher: its first satellite carrying its in-house Trillium AI chips will test whether running AI compute in orbit is feasible.
At OpenAI's 2026 developer conference, DevDay, the company's CEO announced Dots, a new generation of always-on AI agent that runs in the background and proactively advances multi-step tasks.