AI News Weekly: May 18 β May 25, 2026
Google I/O 2026 unveils Gemini 3.5 and agent-first platforms, OpenAI solves an 80-year-old math conjecture and prepares for IPO, while the EU simplifies the AI Act and Standard Chartered cuts 7,000 jobs in an AI-driven restructuring.
AI Weekly: May 18 β May 25, 2026
- Google I/O 2026: Gemini 3.5, Antigravity 2.0, and the Agent-First Platform
- OpenAI Solves an 80-Year-Old Math Conjecture
- OpenAI Prepares for IPO
- White House Postpones AI Executive Order
- EU Simplifies AI Act and Bans Nudification Apps
- Standard Chartered Cuts 7,000 Jobs for AI
- Google and Blackstone Launch AI Cloud Venture
- OpenAI and Dell Partner on On-Premises Codex
- The Capex Arms Race Accelerates
- What to Watch Next
Google I/O 2026: Gemini 3.5, Antigravity 2.0, and the Agent-First Platform
Google's I/O 2026 keynote on May 19 marked a decisive shift from "AI that assists" to "agents that act." The centerpiece was the Gemini 3.5 series, with Gemini 3.5 Flash launching immediately and Gemini 3.5 Pro scheduled for June. Google positioned this as a new model family combining "frontier intelligence with action."
Beyond the models, Google announced Antigravity 2.0, an agent-first development platform with a new CLI for orchestrating specialized subagents. The platform includes cross-platform terminal sandboxing, credential masking, and hardened Git policies. Developers can now spin up subagents to handle complex workflows without manual oversight.
Key developer announcements included:
- Managed Agents in the Gemini API β a single API call provisions a fully sandboxed agent with remote execution
- Antigravity SDK for self-hosting the agent harness on your own infrastructure
- WebMCP β a proposed open web standard (origin trial in Chrome 149) allowing developers to expose structured tools for browser-based AI agents
- Android CLI and skills β enabling any LLM to tap into Android Studio's heavy-lifting capabilities
- Migration agent β a new Android Studio feature that converts React Native, web framework, or iOS code into native Kotlin apps in hours instead of weeks
- Gemini Omni β a new multimodal model enabling high-quality video generation from any input type
Google also shifted the Gemini app from daily prompt limits to a "compute-used" pricing model, factoring in prompt complexity, features used, and chat length.
Sources:
OpenAI Solves an 80-Year-Old Math Conjecture
On May 20, OpenAI announced that an internal general-purpose reasoning model had autonomously disproved a central conjecture in discrete geometry first posed by Paul ErdΕs in 1946 β the planar unit distance problem.
The problem asks: given n points in a plane, how many pairs can be exactly distance 1 apart? For 80 years, mathematicians believed the "square grid" construction was essentially optimal, with the conjectured upper bound of n^(1+o(1)). The OpenAI model produced an infinite family of counterexamples yielding a polynomial improvement β configurations with at least n^(1+Ξ΄) unit-distance pairs for a fixed Ξ΄ > 0 (later refined by Princeton's Will Sawin to Ξ΄ = 0.014).
What makes this result extraordinary:
- It was solved by a general-purpose reasoning model, not a system specifically trained for mathematics
- The proof draws on algebraic number theory β infinite class field towers and GolodβShafarevich theory β creating an unexpected bridge to elementary geometry
- The proof has been verified by external mathematicians, including Fields Medalist Tim Gowers, who called it "a milestone in AI mathematics"
- Number theorist Arul Shankar stated the work demonstrates that "current AI models go beyond just helpers to human mathematicians β they are capable of having original ingenious ideas"
This is the first time a prominent open problem central to an active field of mathematics has been solved autonomously by AI.
Sources:
OpenAI Prepares for IPO
The New York Times reported on May 20 that OpenAI is working with Goldman Sachs and Morgan Stanley to prepare IPO paperwork, with a filing potentially as early as the week of May 25.
Context for the listing:
- OpenAI closed a $122 billion funding round β the largest private venture round in history β pushing its post-money valuation to $852 billion
- Amazon led the round with up to $50 billion, alongside Nvidia ($30B) and SoftBank ($30B)
- Microsoft's initial $13 billion investment has grown to an estimated $228.3 billion in value
- SoftBank's total $64 billion investment is worth approximately $99.3 billion
The IPO timing is being carefully watched against market conditions. OpenAI's path to public markets represents a pivotal moment for the AI industry, transitioning one of the world's most valuable private companies into public scrutiny.
Sources:
White House Postpones AI Executive Order
On May 20, the White House postponed an executive order that would have established a voluntary framework requiring AI companies to share advanced models with the government 14 to 90 days before public release.
The draft order, reported by Axios, called for:
- A voluntary model-sharing framework under which AI labs would provide government access before public release
- Creation of a vault for security vulnerabilities
- Voluntary participation from major AI companies
The postponement reflects internal White House infighting over the scope and approach. Industry reaction has been mixed β some welcome the delay as avoiding premature regulation, while others see it as a missed opportunity to establish trust frameworks.
Sources:
EU Simplifies AI Act and Bans Nudification Apps
On May 7, the European Union reached an "omnibus" deal to simplify the AI Act, balancing innovation with new consumer protections:
Timeline relief:
- High-risk AI compliance deferred by 16 months to December 2027
- Generative AI watermarking requirements moved to December 2, 2026 (only 4-month delay)
- Machinery sector carved out of scope
New prohibitions:
- Total ban on "nudification apps" β AI tools generating non-consensual sexualized images of real people
- Ban covers images, video, and audio, including AI-generated child sexual abuse material
- Takes effect December 2, 2026
The deal was accelerated after fake explicit images of Italian Prime Minister Giorgia Meloni circulated online, creating political pressure for immediate action. The simplification represents a pragmatic recalibration β giving industry more time to comply while tightening protections where public harm is clearest.
Sources:
Standard Chartered Cuts 7,000 Jobs for AI
On May 19, Standard Chartered announced it will eliminate more than 7,000 jobs over the next four years (approximately 15% of corporate roles) as it replaces "lower-value human capital" with AI and automation.
This is one of the most explicit public acknowledgments by a major global bank that AI is driving structural headcount reduction, not just efficiency gains. The cuts span operations across the bank's global footprint, with the bank citing a strategy to pursue higher profitability and operational efficiency through technology.
The announcement has significant implications for the financial services sector, where AI adoption in back-office operations, compliance, and customer service has been accelerating. If other major banks follow suit, the cumulative impact on global banking employment could be substantial.
Sources:
Google and Blackstone Launch AI Cloud Venture
On May 18, Google and Blackstone announced a joint venture to create a new AI cloud company, directly competing with CoreWeave in the burgeoning third-party AI infrastructure market.
Key details:
- Blackstone committed $5 billion in initial equity to bring 500 MW of compute capacity online by 2027
- The venture will be built around Google's custom TPU chips, offering an alternative to Nvidia-dominated infrastructure
- Plans to scale significantly beyond the initial 500 MW
- Google Cloud CEO Thomas Kurian said the JV "helps meet growing demand for TPUs, which are optimized specifically for efficiency and performance in the AI era"
This move signals Google's strategy to loosen Nvidia's grip on the AI hardware market by pairing its in-house silicon with Wall Street's capital. It also reflects the broader trend of Big Tech firms spending over $700 billion on AI infrastructure in 2026.
Sources:
OpenAI and Dell Partner on On-Premises Codex
On May 18, OpenAI and Dell Technologies announced a partnership to bring Codex into hybrid and on-premises enterprise environments through the Dell AI Data Platform and Dell AI Factory.
This addresses a critical gap for enterprises that cannot send sensitive data to public cloud APIs. The partnership enables:
- Deployment of AI coding agents within governed data environments
- Integration with existing enterprise workflows and systems
- Secure agent-based development without data exfiltration risks
The deal was announced at Dell Technologies World in Las Vegas and represents OpenAI's push into the enterprise on-premises market, where data sovereignty and compliance requirements make cloud-only solutions insufficient.
Sources:
The Capex Arms Race Accelerates
This week's announcements paint a picture of unprecedented capital deployment in AI infrastructure:
- Meta raised its 2026 capex guidance to $125β145 billion, nearly double last year's pace, and committed $200 billion to a single AI data center in rural Louisiana
- Amazon is spending $200 billion on AI infrastructure, with management noting it has already secured agreements to utilize the expanded computing capacity
- Google disclosed its capex has grown from $31 billion in 2022 to $185 billion in 2026
- Combined mega-cap spending across Amazon, Microsoft, Google, and Meta totals approximately $710β725 billion for 2026 β a 75% year-over-year increase
The free cash flow impact is already visible: Amazon's free cash flow fell from $38 billion to $1.2 billion as it pours capital into data centers, custom silicon, and GPU procurement. Short-seller Michael Burry has argued the group understates annual depreciation by $176 billion over 2026β2028 by applying five-to-six-year GPU life schedules.
Sources:
What to Watch Next
OpenAI IPO filing: If OpenAI files in the coming weeks, the valuation, governance structure, and profit-sharing arrangements for the original mission will be scrutinized intensely. This could set a precedent for how AI companies transition from private to public.
Gemini 3.5 Pro launch: Scheduled for June, this will be the first real test of whether Google's "frontier intelligence with action" positioning holds up against OpenAI's reasoning models, especially after the ErdΕs conjecture breakthrough.
EU AI Act implementation: With the omnibus deal finalized, the real work of compliance begins. The December 2026 watermarking deadline for generative AI is the nearest cliff edge. Watch for how model providers adapt their output pipelines.
Standard Chartered's playbook: If other major banks follow StanChart's lead with explicit AI-driven headcount targets, we could see a wave of similar announcements across the financial sector in Q2βQ3 2026.
Google-Blackstone TPU cloud: The 500 MW target for 2027 will be a litmus test for whether Google's custom silicon can compete with Nvidia's ecosystem at scale. Success here could fundamentally reshape the AI hardware market.
US AI policy direction: The postponed executive order leaves a regulatory vacuum. Watch for whether the White House returns with a revised approach, or whether AI governance shifts entirely to congressional action.
This report covers verified AI news from May 18β25, 2026. All stories are sourced from primary announcements, reputable news outlets, and official press releases.