AI news, 25 October 2026
Block unveiled Dessa, an AI voice model that processes audio; NVIDIA opened the Nemotron 3.5 Bolt 4 GB model; OpenAI resumed AI training after fixing a Brotli flaw.
19 stories, each labelled: available now, announced, unconfirmed or research. Get them ranked and explained for your product →
Top stories
Block launches Dessa AI voice model that processes audio directly
Block launched Dessa, an AI voice model that engages in speech without converting to text. It processes audio directly and can recognize emotional cues such as laughter and sadness; a small group of AI safety researchers will examine potential risks.
NVIDIA releases open Nemotron 3.5 Bolt E4B 4GB model
NVIDIA released Nemotron 3.5 Bolt E4B 4GB, an open small language model post‑trained from Gemma 4 E4B. It targets about 4 GB of VRAM on RTX GPUs and Jetson devices for local voice assistants, game characters and robots.
OpenAI Resumes Frontier AI Training After Fixing Brotli Vulnerability
OpenAI resumed training its advanced AI models after strengthening security and fixing a Brotli flaw. The company disclosed the fix following earlier vulnerabilities highlighted by METR and Redwood Research.
Pennsylvania Governor signs AI Deceptive Advertising Act requiring ad labels
Pennsylvania Governor Josh Shapiro signed the Artificial Intelligence Deceptive Advertising Act into law, mandating that any ad—still image, video, audio, or text—significantly modified or generated by AI include a label in the same format as the ad itself, applying to TV commercials and printed flyers regardless of where the ad was created.
FilmFlow Converts Language or Video into 3D Camera Trajectories
FilmFlow is an agentic framework that turns natural‑language instructions or reference videos into explicit camera trajectories for dynamic 3D environments, using a Director agent that composes high‑level motions from concurrent atomic operations.