🤖 Top AI News — May 23, 2026
Published today, sourced from stories within the last 24 hours.
1. 🎬 Google Debuts Gemini Omni — An “Anything-to-Anything” AI Model
Google has released Gemini Omni Flash, the first model in its new Omni family of generative AI models. The big promise: turn any kind of input — photo, video, text — into anything else. For now, it focuses on video generation and editing, available through Google’s Flow platform.
Early hands-on testing shows Omni improves on its predecessor Veo with better character consistency and the ability to use both uploaded video and text prompts as starting points. However, results remain inconsistent — AI-generated characters still glitch with sudden orientation changes and object inconsistencies. The model is live now in Flow.
2. 🔍 Google Search AI Breaks on Simple Words Like “Disregard”
Google’s new AI-powered Search experience, which rolled out this week with AI Overviews front and center, is hitting a bizarre edge case: searching for words like “disregard,” “ignore,” or “skip” causes the AI to respond as if you’re dismissing it, rather than looking up a definition.
Type “disregard” and you’d get an AI Overview saying “Got it. If you need anything else or have a new question later, just let me know!” — followed by a huge blank space pushing actual search results off-screen. Google has since stopped showing AI Overviews for the term and says a fix is coming soon.
Source: The Verge | TechCrunch
3. 🕶️ Google’s AI Glasses Are “Almost There” — Hands-On Impressions
At Google I/O this week, TechCrunch got a hands-on with Google’s upcoming Android XR smart glasses — the ones with an in-lens display, not the audio-only pair shipping this fall. The glasses overlay Gemini-powered information directly into your field of view: weather widgets, walking directions, Uber pickup details, live translation, and even custom AI-designed widgets.
The verdict? “Almost there” — promising but still clearly prototype hardware. Google first announced the glasses at last year’s I/O with a $150M commitment alongside Warby Parker.
4. 🔒 Anthropic Opens Up Claude Mythos Preview Security Tools
Anthropic is making the security tools it built for Claude Mythos Preview more widely available. Qualifying customers can now request access to skills, a Claude harness, and a threat model builder as part of an update on Project Glasswing.
The company also published a public dashboard of open-source vulnerabilities disclosed by Mythos Preview and plans to expand Glasswing to additional partners.
5. ✈️ AI Used to Resurrect Voices of Dead Pilots — NTSB Forced to Block Records
In a striking case of AI misuse, people used AI tools to reconstruct cockpit voice recordings from a spectrogram image in the NTSB’s public docket for UPS Flight 2976, a fatal crash in Louisville, Kentucky. The reconstructed audio of deceased pilots’ voices began circulating online.
The NTSB, which is prohibited by federal law from releasing actual cockpit audio, temporarily shut down its entire docket system. Access has since been restored, but 42 investigations remain closed pending review — including the UPS Flight 2976 case.
6. 💰 How VCs and Founders Are Inflating AI Startup ARR
A growing controversy in AI startup land: annual recurring revenue (ARR) figures are being stretched to make AI companies look bigger than they are. Spellbook CEO Scott Stevenson publicly called it a “huge scam,” alleging that major funds are supporting inflated metrics to mislead journalists and score PR coverage.
Traditional ARR measures revenue from active customer contracts, but AI startups are increasingly including non-contractual, consumption-based, or even projected revenue in their ARR figures — a practice their investors are fully aware of.
7. 🚀 SpaceX $1.75T IPO Filing — The AI Infrastructure Play
SpaceX has filed to go public at a staggering $1.75 trillion valuation, and AI is central to the math. The company’s $15 billion/year deal with Anthropic for AI compute capacity on the Colossus supercomputer cluster underscores how SpaceX is positioning itself as critical AI infrastructure, not just a space company.
Published May 23, 2026. All stories sourced from reports published within the last 24 hours. For the latest updates, visit aimasternow.blog.
Leave a Reply