Tech News — Friday, May 22, 2026

Google I/O Reshapes AI, Nvidia Shatters Records, OpenAI Eyes the Smartphone, and Anthropic Finds Thousands of Hidden Vulnerabilities

Four seismic stories that defined the last 48 hours in technology.

🌐 Google I/O 2026: Agents Take Center Stage

AI Google AI technology abstract visualization

Google's annual developer conference has rarely carried as much weight as it did this week. At Google I/O 2026, CEO Sundar Pichai declared the era of AI assistants officially over — replaced by something far more consequential: autonomous agents. The headline release is Gemini 3.5 Flash, a model Google describes as delivering frontier-level intelligence at less than half the cost of comparable models, outperforming the previous Gemini 3.1 Pro on demanding coding and agentic benchmarks while maintaining the speed the Flash line is known for.

Perhaps the most architecturally significant announcement was WebMCP, a proposed open web standard that would allow developers to expose structured tools directly to browser-based AI agents. If adopted broadly, WebMCP could turn the entire web into a programmable surface for autonomous AI — a fundamental shift in how software interacts with the internet. Google is launching an experimental origin trial in Chrome 149, giving developers their first chance to build against the standard.

Google also announced Gemini Spark, a 24/7 personal AI agent designed to operate continuously in the background across phones and laptops, managing digital tasks even when the device's screen is off. Complementing this is a deep Gemini integration into Chrome itself, enabling real-time webpage translation, contextual summarization, and smart form-filling — all without leaving the browser. For researchers, Science Skills offers agentic access to over 30 major life science databases, compressing workflows in genomics and structural bioinformatics from days to minutes.

100+

Announcements at I/O 2026

30+

Life Science Databases in Science Skills

<50%

Cost vs. Comparable Frontier Models

"We've moved from AI that assists to AI that acts — and the question is no longer whether agents will transform software, but how quickly." — Sundar Pichai, Google I/O 2026

📈 Nvidia Posts $81.6B in Revenue — A New Record for Any Chipmaker

Breaking Semiconductors Data center servers representing AI chip infrastructure

Nvidia reported its most extraordinary quarter in company history on Wednesday, posting quarterly revenue of $81.6 billion — an 85 percent increase year-on-year — and a record net profit of $58.3 billion, up more than 200 percent from the same period in 2025. The results, driven almost entirely by insatiable demand for AI infrastructure, sent markets surging and reignited debate about whether the AI spending boom has any ceiling in sight. For context, Nvidia's single-quarter revenue now exceeds the annual revenue of many Fortune 100 companies.

The company's data-center segment — the engine behind its AI dominance — grew 92 percent year-on-year to $75.2 billion, accounting for roughly 92 cents of every dollar Nvidia earned. CEO Jensen Huang struck a bullish tone in his earnings call, guiding the current quarter to $91 billion in revenue — surpassing nearly every analyst estimate. The Blackwell Ultra GPU architecture, shipping at scale since early 2026, was cited as a key demand driver, with hyperscalers and sovereign AI programs across Asia, Europe, and the Middle East ordering at unprecedented volumes.

Not all signals were frictionless. Nvidia separately confirmed it has "largely conceded" China's AI chip market to Huawei, following tightened U.S. export restrictions that blocked sales of its most advanced processors to Chinese buyers. Meanwhile, analysts at IDC note that while Nvidia holds roughly 81 percent of the AI data-center chip market, custom ASIC solutions from Google, Amazon, and Microsoft are gaining ground — with TrendForce projecting 44.6 percent growth in custom ASIC shipments this year versus 16.1 percent for GPUs. The arms race is intensifying, even for its undisputed leader.

$81.6B

Quarterly Revenue

$58.3B

Net Profit (Record)

+85%

Revenue Growth YoY

81%

AI Data Center Market Share

📱 OpenAI's AI Smartphone Is Real — and It's Coming Sooner Than You Think

Devices Breaking Smartphone device representing next-generation AI hardware

For months, reports of an OpenAI-designed smartphone circulated as speculation — but new details this week have hardened the project into something unmistakably real. OpenAI, working in partnership with designer Jony Ive and manufacturing partner Luxshare Precision Industry, is fast-tracking its first AI agent phone toward mass production as early as the first half of 2027. The device's chip strategy is equally ambitious: it will run on a customized version of MediaTek's Dimensity 9600, built on TSMC's cutting-edge N2P node, paired with a dual-NPU architecture designed specifically to handle the heterogeneous compute demands of always-on AI agents.

The core premise of the device diverges sharply from the traditional smartphone paradigm. Rather than organizing user experience around apps, the OpenAI phone is designed around AI agents that replace apps entirely — understanding the user's real-time context across location, activity, communication, and calendar, then executing tasks autonomously. CEO Sam Altman has been direct about his ambitions: "It feels like a good time to seriously rethink how operating systems and user interfaces are designed," he posted this week, signaling that this project is not a side experiment but a core strategic bet for the company.

OpenAI's Chief Global Affairs Officer Chris Lehane confirmed a first hardware announcement is expected in the second half of 2026, with a commercial launch targeting early 2027. That timeline puts OpenAI in direct competition with Apple's own accelerating AI hardware roadmap — and raises a question the industry is only beginning to grapple with: if AI agents can replace apps, what exactly is an operating system for?

"The smartphone remains uniquely positioned for AI agent use because it is the only device that captures a user's full real-time state — location, activity, communication, and context." — OpenAI device research briefing, 2026

🔐 Anthropic's Claude Mythos Found Thousands of Zero-Days — In Every Major OS and Browser

Security Alert AI Cybersecurity digital lock representing vulnerability research

Anthropic this week continued to draw attention to one of the most consequential AI safety demonstrations of 2026: Project Glasswing, a cybersecurity initiative that uses a preview version of its new frontier model, Claude Mythos, to identify previously unknown software vulnerabilities at a scale that no human team could match. Over a matter of weeks in internal testing, Claude Mythos identified thousands of zero-day vulnerabilities — flaws entirely unknown to the software's own developers — spanning every major operating system (Windows, macOS, Linux, FreeBSD, OpenBSD) and every major web browser (Chrome, Firefox, Safari, and Edge).

The scale of the discovery is difficult to overstate. At the time of Anthropic's April announcement, over 99 percent of the identified vulnerabilities remained unpatched — a situation that underscores both the extraordinary capability of the model and the fragility of the software infrastructure that underpins the modern internet. Anthropic has deliberately withheld Claude Mythos from general availability, citing the dual-use danger: the same capabilities that let it find vulnerabilities defensively could, in hostile hands, be turned to exploit them offensively.

Access to the model has instead been extended to a tightly controlled group of partners under Project Glasswing's framework, including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks. Together, these organizations are working to responsibly disclose and patch the discovered flaws before bad actors develop equivalent capabilities independently. Anthropic frames the project as an "urgent attempt" to give defenders a durable structural advantage before the offense-defense balance in AI-driven cybersecurity tips irreversibly toward attackers.

1000s

Zero-Days Discovered

>99%

Unpatched at Time of Disclosure

Major Tech Partners in Program

"Project Glasswing is an urgent attempt to employ frontier model capabilities for defensive purposes before those same capabilities are adopted by hostile actors." — Anthropic, April 2026

Search This Blog

GADGETGLOW BYTES