Waymo Halts Robotaxis in Five Cities After Flood Patch Fails, No Fix in Sight

Waymo expanded its robotaxi service pause to five U.S. cities on May 21 after a software patch pushed to its entire 3,791-vehicle fleet failed to prevent autonomous vehicles from driving into standing water. The expanded pause covers Atlanta, Dallas, Houston, San Antonio, and Austin — the latter two already suspended before the patch rollout. A Waymo vehicle in Atlanta was documented driving into a flooded street and became stuck for roughly an hour. The company acknowledged it has no permanent fix for the problem, marking a critical setback for the world’s most commercially advanced robotaxi fleet, which now cannot reliably operate during heavy rain.

Sources: TechCrunch

Hark Raises $700M Series A to Build Universal AI Interface and Dedicated Hardware

Hark, an AI lab founded by Brett Adcock, creator of robotics startup Figure.AI and electric aircraft company Archer Aviation, raised $700 million in a Series A round valuing the company at $6 billion on May 21. Led by Parkway Venture Capital, the round included strategic investors NVIDIA, AMD Ventures, Intel Capital, Qualcomm Ventures, and Salesforce Ventures. Hark is building a secretive universal AI interface and plans to release its first multimodal models this summer, followed by dedicated hardware devices. With $800 million raised total and virtually no public product disclosures, the company has become one of the most closely watched and deliberately opaque AI ventures of 2026.

Sources: TechCrunch

NTSB Shuts Down Public Docket After AI Reconstructs Voices of Crash Victims

The National Transportation Safety Board temporarily shut down its public docket system on May 22 after discovering AI tools had been used to reconstruct the voices of pilots killed in a UPS cargo plane crash. Using publicly available cockpit voice recorder transcripts and spectrograms, anonymous individuals created AI voice approximations with tools such as Codex that began circulating online. The agency restored public access by week’s end but kept 42 active investigations sealed pending review. The incident prompted calls for federal guidelines on AI reproduction of the deceased, with victim advocacy groups questioning government obligations to protect data in federal safety records.

Sources: TechCrunch

SpaceX IPO Filing Reveals xAI Burns Natural Gas, Not Solar, to Power Its Data Centers

A SpaceX IPO filing released the week of May 19 revealed that xAI, Elon Musk’s AI company, powers its data centers using dozens of unregulated natural gas turbines and has contracted to spend $2.8 billion on additional gas capacity. The filing shows xAI purchased just $697 million in Tesla Megapacks over two years and has not acquired meaningful solar capacity from Tesla Energy. SpaceX’s filing separately pitches space-based solar as a future alternative, claiming orbital arrays generate more than five times the energy of terrestrial panels. The disclosure drew criticism from climate researchers, who noted the strategy directly contradicts Musk’s longstanding public positioning around clean energy.

Sources: TechCrunch

AI Security Is Failing in Real Time as Attack Speeds Drop to 22 Seconds

The enterprise AI security landscape is deteriorating faster than organizations can respond, according to a TechCrunch analysis published May 24. Researchers at Aikido Security found that compromised Google API keys remain active for up to 23 minutes after deletion as revocations propagate gradually through Google’s infrastructure, a window attackers have exploited to exfiltrate Gemini conversation data and cached files. Industry-wide, the average time between an initial breach and a follow-on attack has dropped from eight hours to 22 seconds, driven largely by AI-powered offensive tools. LinkedIn’s chief information security officer Lea Kissner said she does not expect the industry to reach a stable, sustainable AI security posture for several years.

Sources: TechCrunch

Google’s New AI Search Breaks When You Type ‘Disregard’ — And May Be Doing It on Purpose

Google’s AI-first search redesign, launched at I/O 2026, has produced an unintended outcome: the word “disregard” appears to no longer function as a standard search query. TechCrunch reported on May 22 that typing the term — commonly used in prompt injection attacks to override AI instructions — now returns a broken or empty interface in Google’s new AI-powered search box. The behavior suggests Google is actively filtering prompt injection keywords from its redesigned search layer, which foregrounds Gemini AI summaries and has substantially reduced visibility for organic web links. Google has not publicly confirmed or denied that the filtering is intentional.

Sources: TechCrunch

AI Startup CEO Calls Out Sector-Wide Revenue Inflation as a ‘Huge Scam’

Scott Stevenson, CEO of legal AI startup Spellbook, published a public critique on May 22 alleging systemic revenue misrepresentation across AI startups. Stevenson argued that many AI companies report annualized recurring revenue figures by multiplying a single month of revenue by twelve, regardless of whether they hold subscription contracts guaranteeing those future payments. He called the practice a “huge scam” and accused major venture funds of knowingly amplifying the figures for marketing purposes. The allegation was notable because Stevenson is himself an AI startup founder with a commercial stake in the industry, making him an unusual insider critic at a time when scrutiny of AI startup valuations is intensifying.

Sources: TechCrunch

What's Trending in Tech