Full Summary
This Saturday morning, May 9th, the White House is eyeing mandatory government vetting for advanced AI models before public release, a move confirmed by both Let's Data Science and AOL.com. This comes as the Commerce Department expands its safety testing to include new AI models from Google, Microsoft, and xAI, with these companies voluntarily submitting their models like Google's Gemini, Microsoft's CoPilot, and xAI's Grok for evaluation by the Center for AI Standards and Innovation (CAISI). CAISI has already completed over 40 model evaluations. In other major AI developments, OpenAI launched GPT-Realtime-2 on May 7th, a voice model that reasons in real-time, directly within audio conversations. OpenTools reports it brings GPT-5 level intelligence to live interactions with a massive 128,000 token context window. This model, priced at about $0.048 per minute, saw Zillow increase call success rates from 69% to 95%. OpenAI also released GPT-5.5-Cyber, an AI specifically for cybersecurity defenders, scoring 81.9% on the CyberGym benchmark. Meanwhile, Meta has launched Muse Spark, its first completely closed AI model, moving away from its open-source approach, according to Bitget. This AI, from Meta Superintelligence Labs, handles text, images, and voice, and includes a "Contemplating" mode. DeepSeek, a Chinese AI company, is nearing a $45 billion valuation as China's state-backed "Big Fund" leads a new financing round, Magzter reports. DeepSeek also launched its V4 AI model, which performs exceptionally well with chips made in China, potentially boosting demand for domestic AI chips, The Star confirms. On the hardware front, Futurism reports Google Chrome has been secretly installing a 4-gigabyte Gemini Nano AI model, "weights.bin," on users' computers without consent, leading to privacy concerns and accusations of "malware." If deleted, Chrome re-downloads it. This rapid advancement and increasing government oversight mean new AI tools are being integrated into daily life, from your phone to national security, often without explicit consent or prior testing for public safety.