Reka AI's Rho-1 Runs Text, Video and Robots in One Model

Reka AI previews Rho-1, a 19B omni-model that also drives robots

Reka AI has released a research preview of Rho-1, a 19-billion-parameter omni-model that processes and generates text, images, video and robot control actions inside a single neural network. Rather than routing tasks to specialized models, Rho-1 treats every modality as tokens in one shared context window, with no tool calls or external models. It generates continuous video in real time and adapts to new instructions without restarting, and the same weights that predict camera images also drive robot movements. To offset scarce robot training data, Reka AI built an inverse dynamics model that extracts control signals from ordinary internet videos. Rho-1 trained on 320 H100 GPUs over roughly three months.

Reka AI’s omni-model Rho-1 handles text, images, video, and robot control in a single model →

Open-source AI agent ARTEX breached seven South Korean banks

South Korea activated a round-the-clock cybersecurity emergency after an AI-powered attack tool breached seven financial institutions over about a week from September 28, 2026, exposing data on more than 65,000 customers. The tool, ARTEX AI, is an autonomous penetration-testing system built on large language models that handles reconnaissance, attack and verification without continuous human direction. It was a winning entry in a challenge run by Baidu’s Security Response Center and distributed openly on GitHub. The method was credential stuffing with adaptive execution, targeting auxiliary systems such as loan agent portals rather than core transaction infrastructure. The national cyber alert level rose from Caution to Watch, and KISA issued advisories to 28,000 companies.

Open-Source AI Agent Hacked Seven South Korean Banks, Exposing 65,000 Records →

Quinnipiac poll: 77 percent of Americans want AI slowed or halted

A Quinnipiac University poll found 47 percent of Americans want AI development to slow down and 30 percent want it stopped entirely until safety is verified, while just 5 percent want it to proceed faster. Eighty-six percent support independent safety standards under debate for AI companies even if those standards slow progress. Seventy-three percent worry AI could eventually threaten humanity’s survival, and 74 percent have little or no trust in AI company leaders. Fifty-three percent say AI will do more harm than good in daily life, 91 percent call US guardrails important, and 72 percent oppose an AI data center in their own community. The poll surveyed 1,202 US adults from September 24 to 27, with a margin of error of plus or minus 3.5 points.

Most Americans want AI development to slow down or stop entirely, new poll finds →

Aleph Alpha releases Kolibri, a 78B open-weight German-English model

Aleph Alpha has released Kolibri, a German-English language model with 78 billion parameters, about three billion active per token through a mixture-of-experts architecture. The company claims Kolibri sits on the Pareto front of quality and operating cost in both languages, outperforming compared models with similar architectures, some significantly older, on either metric. German accounts for 21.3 percent of training data, backed by a dedicated German data pipeline; Chinese models were used to generate synthetic training data. Kolibri scores 71 percent on German benchmarks while decoding faster than GPT OSS A5B, Qwen 3.6 A3B and Gemma 4 A4B. It supports context windows up to one million tokens, trained on 768 B200 GPUs in Germany and Finland, and ships under an Apache 2.0 license on Hugging Face.

Aleph Alpha releases Kolibri, an open-weight model that makes the case for European AI sovereignty →

Study: female-presenting AI agents paid 10 percent less than male ones

A University of Limerick study titled Human-Like and Male? How AI Assistant Design Relates to Trust and Monetary Reward at Work in VR found that AI agents presenting as female would be paid 10 percent less than male-presenting counterparts for identical work. Researchers placed 189 human workers in a virtual reality office with four assistants: a text chatbot, a desk robot, and two human-like agents named Johan and Johanna, all running the same underlying model, OpenAI’s gpt-4-1106-preview. Participants rated Johan and Johanna similarly, yet both men and women paid Johanna significantly less. Most participants said in interviews they would not treat male- and female-presenting agents differently. The findings arrive as almost 79 percent of companies reported using AI agents this year.

Female AI agents would get paid 10% less than male ones, study finds →