AI News Weekly Summary — May 11–17, 2026 | Gemini Intelligence, Uber vs Waymo, Anthropic $30B

This AI news weekly May 11–17, 2026 covers: Google reveals Gemini Intelligence on Android and the first generation of “Googlebooks”. In parallel, it blocks the first observed AI-generated zero-day exploit. Uber commits $10B+ to Waymo alternatives.

Per Anthropic’s CEO, the company hit a $30B annualised revenue run rate — described as 80x growth on an annualised basis. OpenAI launches a Deployment Company backed by $4B+.

1. Google Android Show I/O Edition — Gemini Intelligence, Googlebooks, Agentic Android

On May 12, Google held its Android Show: I/O Edition — a pre-show ahead of next week’s Google I/O keynote. Headline: Gemini Intelligence, an agentic AI layer that “reads what’s on your screen, moves across apps, and completes multi-step tasks without constant prompting.”

Announcement pack:

• Googlebooks — first laptops designed from the ground up for Gemini Intelligence, launching this fall.
• Android Auto — refreshed UI in Material 3 Expressive style for in-car displays.
• Gemini in Chrome — rolling out to select Android 12+ devices in the US from late June.
• Create My Widget — natural-language widget generation (e.g., “weather, only wind and rain”).

Google is racing Apple, which is launching the rebuilt Siri at WWDC in June. The stakes are which AI vendor owns the smartphone’s system layer.

2. Google Blocks the First AI-Generated Zero-Day

On May 11, Google’s Threat Intelligence Group disclosed it thwarted a “mass exploitation event” — an attempt by a criminal group to use AI to find and exploit a previously unknown vulnerability. John Hultquist (Google) confirmed that hackers are using agentic tools such as OpenClaw to autonomously hunt zero-days.

Per Google, the most serious observed case involved bypassing two-factor authentication. On the same day, Anthropic, OpenAI, and Google publicly highlighted the rising sophistication of agentic attacks.

This is the moment cybersecurity experts have been warning about for years: offensive AI in the hands of criminal groups, not just state actors. Context of the week: Trump is reportedly considering an executive order requiring pre-release safety testing of frontier models, modeled on the Food and Drug Administration (FDA) drug-review process.

3. Anthropic at $30B Annualised Run Rate — 80x Growth on Annualised Basis

On May 11, Anthropic CEO Dario Amodei revealed the company hit a $30B annualised revenue run rate in Q1 2026 — described as 80x growth on an annualised basis. Trajectory (all values in USD as run rate): 87M in January 2024, 1B by end of 2024, 9B by end of 2025, 14B in February 2026, 19B in March, 30B in April.

Primary engine: enterprise. Over 1,000 customers now spend $1M+ per year on Claude. Over half of Claude Code revenue comes from enterprise. Customers include Netflix, Spotify, KPMG, L’Oréal, and Salesforce.

By these numbers, Anthropic has passed OpenAI in reported enterprise adoption and in coding. The revenue claim itself needs care — these are annualised run rates, not closed full-year revenue. Two monetization models are running in parallel: Anthropic “for business,” OpenAI “for everyone with a credit card.”

4. Uber Commits $10B+ to Robotaxi Alternatives to Waymo

On May 15, Electrek and FT reported that Uber is investing $10B+ in autonomous fleets without Waymo.

Breakdown:

• $1.25B in Rivian — up to 50,000 autonomous R2 robotaxis, San Francisco and Miami launch 2028.
• Increased investment in Lucid to $500M — Lucid Robotaxi fleet expanded to 35,000 vehicles (prior: 20,000).
• Added: a partnership with Nuro — an autonomous-vehicle tech provider just cleared by California to test Lucid Gravity vehicles without a safety driver.

In parallel: Uber executives are publicly criticizing Waymo — CTO Praveen Neppalli shared a “scary” Waymo encounter clip in San Francisco in April. Waymo still operates on Uber’s platform in Austin and Atlanta.

Uber is openly opting out of being a passive distributor. With $10B+ and two vehicle makers (Rivian, Lucid) plus an autonomy partner (Nuro), it’s building a parallel stack. Waymo is the leader today (1,400 sq mi, 11 cities), but it now has a competitor with the distribution Tesla lacks. Tesla, which entered Dallas and Houston 7 weeks behind Waymo, gets another competitor in the robotaxi segment.

5. OpenAI Deployment Company — $4B+ to Push Enterprise Adoption

On May 11, OpenAI announced the launch of the OpenAI Deployment Company — a separate entity with $4B+ initial funding tasked with helping enterprises with AI deployment through “Forward Deployed Engineers” and consulting services.

First acquisition: Tomoro — an AI consultancy with ~150 AI engineers and deployment specialists. In parallel, Anthropic is running its own “deployment vehicle” initiative with roughly $1.5B in initial funding.

OpenAI is shifting from a pure product model toward enterprise deployment supported by consulting. The “client self-onboards after signing the SaaS contract” model is ending for advanced agents.

6. Waymo Expands to 1,400 Square Miles Across 11 Cities

On May 13, Waymo announced expanding its service area by about 27% — to over 1,400 square miles across 11 US cities (more than the area of Rhode Island). Fleet: roughly 3,000 robotaxis. Target: 1M rides per week by end of 2026, on top of 20M+ rides already completed.

A darker note from the same week: on May 12, FOX 4 Dallas published a video of a Waymo vehicle running a red light at a busy intersection. Waymo’s response: the signal appeared “heavily dimmed” from the vehicle’s perspective. Federal regulators are already investigating similar incidents (school-bus stop arms ignored, a pedestrian struck in California).

7. OpenAI Daybreak — Frontier AI on the Defensive Side

On May 12, OpenAI launched Daybreak — a cybersecurity initiative using frontier models to detect and patch vulnerabilities before attackers find them. A direct answer to Google’s previous-day announcement of the blocked AI zero-day.

Positioning: softer than Anthropic’s Mythos (locked into Project Glasswing for a chosen few) and GPT-5.5-Cyber in Trusted Access for Cyber (TAC) for verified defenders. Daybreak opens access more broadly, with usage monitoring.

8. ChatGPT — More Web Images for Free Users

On May 12, OpenAI rolled out inline web images for Free users in ChatGPT on web and iOS, for GPT-5.5-Instant responses. Topics: well-known people, places, products. Goal: reduce the Google Images and Wikipedia advantage in “visual answers.”

A direct fight for the same surface targeted by Gemini Intelligence and Perplexity.

9. Anthropic Claude for Small Business — 10-City US Tour, Kicking Off in Chicago

On May 14, Anthropic launched Claude for Small Business — a toggle within Claude Cowork giving access to 15 pre-built agentic workflows (finance, operations, sales, marketing, HR, customer service). Accompanied by a 10-city US tour with free half-day AI training workshops for small businesses, kicking off May 14 in Chicago.

After the enterprise segment (Netflix, Spotify, KPMG), Anthropic is moving down-market. Competition: Microsoft Copilot for Small Business, Google Workspace AI.

10. OpenAI Codex — “From Anywhere”, Coding Agent Update

On May 14, OpenAI announced an update to Codex — its agentic coding tool — allowing sessions to be launched from any environment, not just OpenAI’s web UI. The same day, OpenAI described improvements in sensitive-conversation context recognition in ChatGPT (suicide, self-harm, mental-health crises).

Codex competes directly with Cursor (in talks for a $2B round), Devin from Cognition, and Anthropic’s Claude Code.

11. Anthropic — Credit Limits for Third-Party Agent Tools

This week Anthropic introduced new restrictions and credit limits for Claude subscribers using third-party agent tools. The reason: the rising computational cost of autonomous workflows (multi-hour sessions, dozens of tool calls).

Signal: the “flat-rate subscription” model doesn’t scale economically for agents. Expect the entire market to shift toward measurable consumption pricing (per token, per hour).

12. IBM Think 2026 — watsonx Orchestrate as an Agent Control Plane

On May 5, IBM at Think 2026 announced its broadest enterprise AI toolkit to date, with continued communication in the May 11–17 window. Centerpiece: next-gen watsonx Orchestrate (private preview) — an “agentic control plane” enabling deployment of agents from any source with consistent policy enforcement.

In the bundle: IBM Bob — an agentic dev partner reaching general availability (GA); Context in watsonx.data (private preview) — a federated context layer over business data; IBM Sovereign Core — operational independence.

13. Recursive Superintelligence — $650M at a $4.65B Valuation

On May 13, Recursive Superintelligence (founded in 2025 by former leaders from OpenAI, Google DeepMind, Meta AI, Salesforce AI, and Uber AI) closed a $650M round at a $4.65B valuation. Co-led by GV and Greycroft, with participation from AMD Ventures and NVIDIA. Goal: self-improving AI systems; public launch in mid-2026 with infrastructure in San Francisco and London.

Also: Hint AI (Martha Stewart, AI home management) — $10M seed round led by Slow Ventures (May 13).

14. First US Move Toward Pre-Release Model Regulation

May 5, but echoing through the full week: the Center for AI Standards and Innovation (CAISI) at the US Department of Commerce signed agreements with Google DeepMind, Microsoft, and xAI for access to unreleased models before launch for national-security testing. Earlier CAISI agreements with OpenAI and Anthropic (2024) have been renegotiated. Trump is reportedly considering an executive order requiring testing on the FDA model.

The catalyst: Anthropic’s Mythos. Per Fortune and Federal News Network, the Trump Administration — which previously criticized Biden’s AI safety policies — is now reportedly considering a similar approach (without using the “AI Safety Institute” name).

15. Security — Instructure/Canvas, Jenkins, Android Intrusion Logging

Three smaller security stories this week:

• Instructure (Canvas) — on May 12 reached a settlement with extortion group ShinyHunters after a network breach.
• Jenkins — Checkmarx confirmed on May 11 that a modified Jenkins AST plugin was published to the Jenkins Marketplace (supply chain).
• Android Intrusion Logging — on May 13 Google released an opt-in feature in Advanced Protection Mode for collecting forensic logs to analyze sophisticated spyware.

16. Tesla AI5 — Progress Toward Mass Production in H2 2027

Tesla completed tape-out of AI5 — its next-generation inference chip. Musk claims AI5 delivers roughly 8x compute, 9x memory, 5x bandwidth vs AI4. Musk claims a single AI5 is comparable to NVIDIA H100 for Tesla’s workloads; two chips together are claimed as competitive with Blackwell-class — but at lower cost and power. Engineering samples — late 2026, high-volume production — H2 2027.

Cybercab off the line in Austin in late April, Optimus moving to production. $20B of 2026 CapEx covers, among other things, the Terafab project — Tesla’s own semiconductor fab.

17. Other Important Signals of the Week

• Anthropic Managed Agents — in the May 11–17 window: Dreaming in research preview, while Outcomes, Multiagent Orchestration, and Memory move into public beta on the Claude platform.
• OpenAI “Parameter Golf” — May 13: a challenge to train the best possible language model within a 16 MB weight budget and a short training budget.
• Tech layoffs YTD — over 92,000 tech-sector layoffs since the start of 2026, year-to-date (Storyboard18); Meta confirmed 8,000 cuts for May 20, more in H2 2026.
• Anthropic vs OpenAI — VentureBeat: Anthropic has passed OpenAI in reported enterprise adoption; three threats could erase that lead (compute constraints, no consumer brand, the Pentagon exclusion).
• Sierra — May 4 (before our week, but echoed in press through May 11+) — $950M round at $15B+ valuation.

Weekly Takeaway

The week of May 11–17, 2026 set three big axes. First — the battle for the system layer. Google launched Gemini Intelligence on Android and previewed Googlebooks; Apple is preparing the Siri rebuild powered by Gemini for WWDC. Whoever owns the OS and browser sets the next 5 years of AI distribution. ChatGPT responds with inline images, Perplexity is fighting for the same surface.

Second — AI in cyber leaves the theoretical phase. Google discloses the first observed AI-generated zero-day; OpenAI launches Daybreak as a defensive program with frontier capabilities. Trump is reportedly considering an executive order requiring pre-release model safety testing, modeled on FDA review. The era of “AI safety is Anthropic’s thing” has shifted to “AI safety is a regulatory requirement.”

Third — monetization is polarizing. Anthropic posts a $30B annualised run rate with an enterprise emphasis; OpenAI in the same week launches its Deployment Company with $4B+ for direct deployments. Uber commits $10B+ to its own robotaxis; Waymo scales to 1,400 square miles. Recursive Superintelligence raises $650M on self-improving models. Anthropic introduces credit limits for third-party agents — the first sign that flat-rate subscriptions can’t cover the cost of autonomous workflows.

Full Recaps — April and May 2026

• Week of April 27 – May 3, 2026 — Big Tech Q1, Manus blocked, Pentagon vs Anthropic
• Monthly Recap: April 2026 — Mythos, Opus 4.7, GPT-5.5, $65B for Anthropic

Źródła / Sources (wspólne dla obu wersji):

• Google / Android: Everything announced at Android Show — Engadget, Android Show I/O Edition — TechCrunch, Google races to put Gemini at Android center — CNBC
• Cyberbezpieczeństwo / AI zero-day: Google issues dire warning — Fortune, Google thwarts mass exploitation event — CNBC, First AI-Generated Zero-Day Exploit — SecurityWeek
• Anthropic 30 mld USD / Claude: Anthropic hits $30B run rate — VentureBeat, Anthropic hits $30B run rate — PYMNTS, Anthropic vs OpenAI enterprise — VentureBeat, Anthropic small business tour — SiliconRepublic
• Robotaxi: Waymo expands coverage — Electrek, Uber $10B+ Waymo alternatives — Electrek, Uber $10B robotaxi push — Benzinga, Waymo red light Dallas — FOX 4
• Regulacje: Trump embraces AI oversight — Fortune, Trump moves into AI oversight — CNBC, Government tests AI models — CNN
• Rundy finansowania / inne: Recursive Superintelligence $650M — TechFundingNews, Martha Stewart Hint AI — Fortune, Sierra $950M — TechCrunch, IBM Think 2026 — IBM Newsroom, Tech layoffs 92K — Storyboard18*