Open-Source Voice Model "Audio Interaction" Enables Real-Time Conversation
Existing voice models such as GPT-4o and Qwen3.5-Omni generally wait for the recording to finish before generating a response. The newly released "Audio Interaction," however, takes a completely different approach: it listens continuously in 0.4-second intervals and autonomously decides whether to speak or stay silent.
It can reportedly handle not only translation, transcription, and chat, but also everyday noises like coughs, all in a single stream. The code, model weights, and download instructions are available on GitHub under the Apache 2.0 license, and the training data is expected to be released at a later date.
Unlike conventional turn-based conversational models, this "always listening while deciding" design could significantly expand the range of applications for real-time voice AI.
UK Bank Bans Cyber AI Tool Mythos — OpenAI Proposes an Alternative
It has been reported that after a UK bank was barred from using the cybersecurity AI tool "Mythos," OpenAI stepped in with a proposal for a competing alternative solution.
According to the BBC, the restrictions on Mythos are believed to stem from security concerns, although the details remain unclear. OpenAI's swift move to pitch an alternative highlights how competition among AI companies in the cybersecurity space is intensifying.
For financial institutions, selecting cybersecurity tools comes down to balancing regulatory requirements against real-world effectiveness — an increasingly critical challenge.
Microsoft Patents a Built-in Windows Toggle for Controlling AI Data Collection
Microsoft has been granted a patent for embedding a toggle in Windows that switches AI data collection on and off.
The patent would let users directly control AI data collection at the OS level. As AI services become ubiquitous, concerns over how user data is collected and used continue to grow, and this kind of control feature could well become standard.
That said, a patent does not imply implementation — how, or whether, this will actually be built into Windows remains unknown at this point.
Anthropic Aids NSA Hackers; Meta's AI Bots Also Exploited for Instagram Takeovers
According to WIRED's weekly security roundup, Anthropic's involvement in supporting NSA (National Security Agency) hackers has been making waves. The details are left to WIRED's full article, but the cooperation between AI companies and intelligence agencies remains a subject of ongoing debate.
The same article also reports on a technique in which hackers abuse Meta's AI bots to take over Instagram accounts.
The AI Layoff Trap and the UBI Economic Model — Is Universal Basic Income No Fix for AI Unemployment?
An article titled "The AI Layoff Trap and its UBI economic model" has been gaining traction on Hacker News. As AI-driven layoffs accelerate, the article offers a critical economic analysis challenging the view that universal basic income (UBI) is a cure-all.
With optimism and pessimism mingling over AI's impact on the labor market, the argument that UBI alone cannot solve structural employment problems is a key point of contention. What's needed now is a deeper debate grounded in actual economic models.