Google breaks on one word, Waymo breaks on water, Trump breaks on AI rules
Google's AI search can't handle the word 'disregard', Waymo pulls freeway service, and Trump kills an AI security order under pressure from Musk and Zuckerberg.
Subscribe free →Type "Disregard" Into Google. Watch It Fall Apart.
Google's AI Overviews have a new, embarrassing bug: searching for the word "disregard" causes the system to interpret it as an instruction to ignore its own context, spitting out a chatbot-style non-answer instead of search results. It's a textbook prompt injection failure baked directly into the world's most-used search engine.
Why it matters: This isn't just a quirky glitch — it's a live demonstration of why bolting LLMs onto critical infrastructure without robust input handling is genuinely risky.
Musk and Zuckerberg Talked Trump Out of an AI Security Order
Trump scrapped a planned executive order that would have required pre-release government security reviews of AI models, citing concerns it could slow US competitiveness against China. Reports indicate Musk and Zuckerberg lobbied directly against the order, and Trump obliged — saying he didn't want to "get in the way" of America leading.
Why it matters: The US just signaled that AI safety reviews are optional when tech billionaires push back hard enough, setting a precedent that will echo through every future AI governance debate.
Waymo Is Having a Very Bad Week
Waymo suspended freeway driving across all US markets after its robotaxis struggled to navigate construction zones, then separately paused operations in Atlanta and San Antonio after vehicles drove into flooded streets. Two separate edge-case failures, same week, four cities affected.
Why it matters: Waymo is the gold standard in autonomous vehicles — if it's pulling back this publicly and this broadly, it's a reminder that "almost ready" and "ready" are very different thresholds.
Spotify and Universal Just Legitimized AI Covers — With a Revenue Split
Spotify is partnering with Universal Music Group to let Premium subscribers generate AI covers and remixes of songs, with artists who opt in receiving a share of the revenue. It's the first major label deal to formally license AI music creation at consumer scale.
Why it matters: This is the music industry choosing to monetize AI rather than litigate it out of existence — a model that could reshape how fan creativity, artist rights, and streaming economics interact.
Google's AI Glasses Are Impressive and Not Quite Ready
TechCrunch got hands-on with Google's prototype Android XR glasses, which use Gemini to overlay real-time translation, navigation, and contextual info in your field of view. The demos were genuinely useful — but the hardware is still prototype-grade and the "almost there" verdict echoes Meta's Ray-Ban journey.
Why it matters: Google is the only company with the AI infrastructure, Maps data, and hardware partnerships to make ambient AI glasses actually work — when these ship, they'll be a serious product.
OpenAI Plants Its First International AI Lab in Singapore
OpenAI announced it will open its first Applied AI Lab outside the US in Singapore, backed by a S$300 million commitment through a partnership with Singapore's Ministry of Digital Development. The initiative, called OpenAI for Singapore, was unveiled at the ATx Summit.
Why it matters: OpenAI is building geopolitical infrastructure, not just products — planting labs in strategically important countries signals a race to lock in government partnerships before rivals do.
Quick Hits
- →An AI-written story apparently won a regional slot in the prestigious Commonwealth Short Story Prize — and the literary world has no good process for dealing with it. The Verge
- →Audio generation app Huxe, founded by ex-NotebookLM developers, is shutting down and pulling its apps from both stores. TechCrunch
- →HMD is pre-loading Indian AI chatbot Sarvam's Indus app — supporting 22 Indic languages — on a new smartphone targeting the Indian market. TechCrunch
- →Hacker group TeamPCP is poisoning open source repositories on GitHub at what researchers are calling "an unprecedented scale." Ars Technica
AI TLDR