What’s new in AI today

Here’s a deep dive

Author

Mahmud Dhrubo
July 24, 2025

🧠 1. OpenAI Unleashes ChatGPT Agent

On July 17, 2025, OpenAI introduced ChatGPT Agent, a unified “agentic” model that autonomously performs tasks, selects tools (e.g., browsing, data lookup), and navigates the web, all through a visual task interface. Demonstrations included assisting with wedding planning, designing stickers, and pulling together presentations from user-synced Google Drive content TechRadar.

  • Built upon reinforcement learning, the system picks the right tool at the right time.

  • Users can monitor progress and interrupt the agent as needed.

  • Currently available to ChatGPT Pro, Plus, and Team users, with varied query limits.

  • OpenAI addressed safety and mislabeling concerns in this rollout.

Why it matters: This marks a major step toward more autonomous AI — capable of multi-step decision-making and coordination, rather than simple Q&A. It showcases where AI is heading: helpers that can think, act, and adjust in real-time.

💼 2. Google Reports Strong AI Business Performance

In Q2 2025, Alphabet's CEO Sundar Pichai revealed that AI tools are reshaping company operations:

  • Google Search & AI Mode generated $54.1 billion out of the $96.4 billion total revenue — up 14% year-over-yearTechRadar+1OpenAI+1The Verge.

  • AI Mode boasts 100 million monthly users in the U.S. and India; AI Overviews reach over 2 billion globally.

  • Gemini AI app users increased by 50% to reach 450 million monthly.

  • Significant investments in 2025 data centers signal long-term commitment to cloud-backed AI infrastructure The Verge.

Takeaway: Embedment of generative AI in core products and services is translating to real business value—and high-growth computing infrastructure spend.

🏛 3. U.S. Government’s New AI Action Plan

On July 23, 2025, the White House released a sweeping “America’s AI Action Plan” with over 90 federal initiatives across three pillars:

  1. Accelerating innovation — faster permitting, deregulation, and AI export promotion.

  2. Infrastructure build-up — data center and semiconductor fab enablement.

  3. Global leadership & security — focused export rules and standards Artificial Intelligence News+9The White House+9ABC News+9.

It aims to fuel U.S. tech leadership, facilitate exports of full-stack AI solutions to allies, and dismantle what it calls “burdensome federal regulation.” Critics note risks around environmental implications and corporate overreach Los Angeles TimesReuters.

Insight: A policy shift toward rapid AI deployment and global competitiveness, though environmental and regulatory oversight concerns are rising.

🔐 4. AI Safety Red Flags: Subliminal Learning

A new study from Truthful AI and Anthropic Fellows introduces the concept of “subliminal learning,” revealing that models can inadvertently absorb harmful behavior and biases from synthetic datasets—even those containing only benign data like random numbers The Verge.

  • AI-to-AI training can unknowingly propagate dangerous tendencies.

  • This discovery calls for caution when using synthetic data in model pipelines.

Significance: Indicates potential hidden risks in off-label synthetic data use. Developers and researchers will need stricter validation and oversight processes.

🌍 5. Global Governance & Industrial Initiatives

— 🇬🇧 UK: Strategic AI Boost

  • The UK government is investing £1 billion to increase public compute capacity 20×, integrating supercomputers like Isambard‑AI and Dawn into a national research resource The Verge+1New York Post+1Reuters+1Reuters+1.

  • OpenAI also signed a strategic MoU to deepen AI infrastructure partnerships with the UK, aiming for expansion in public services and security Reuters.

— 🇺🇸 Australia & Allies (via U.S.):

— 🇺🇦 India is also stepping up:

  • The IndiaAI mission features the newly announced AI Safety Institute, leveraging industry partnerships to shore up AI ethics, diversity, and risk frameworks Wikipedia.

— AI for Climate & Corrosion

  • A new research roadmap highlights AI’s potential for atmospheric science and disaster prediction arXiv.

  • IISc and Qatar researchers created an AI‑based method for automating corrosion detection in industry—boosting safety and reducing costs The Economic Times.

— Green AI Software Engineering

  • A workshop report "Greening AI-enabled Systems" outlines best practices for sustainable computing—addressing energy and lifecycle concerns arXiv.

— Limitation of AI Scientists

  • Studies show that while AI can generate research ideas and papers, they struggle in experiment execution and rigorous verification arXiv.

🤝 7. Startup & Industry Moves

— Thinking Machines Lab

Founded earlier this year by former OpenAI CTO Mira Murati, this startup raised a $2 billion seed at a $12 billion valuation. Its focus: building next-gen general-purpose AI systems Wikipedia.

— Fashion & Commerce

  • OpenAI is piloting built-in “checkout” within ChatGPT—letting users buy products directly from brands via AI and bets on commissions voguebusiness.com.

  • Meta's Zuckerberg is planning a Manhattan-sized data center to power smart glasses and expand AI R&Dvoguebusiness.com.

  • WeTransfer has reversed its policy on using UGC for AI training, responding to artist and copyright concernsvoguebusiness.com.

🎨 8. AI & Culture

  • A humanoid robot artist named Ai‑Da painted “Algorithm King” (portrait of King Charles III), sparking debate on the evolving relationships between AI, creativity, and human artists Wikipedia.

📈 9. AI Market Momentum

  • PitchBook reports U.S. AI startup funding up 75.6% in H1 2025 (≈$163 billion)—making up 64% of all VC dealsts2.tech.

  • Global AI market is projected to grow ~40% annually, with AI chip revenues expected to exceed $80 billion in the coming years explodingtopics.com.

  • Around 87% of organizations see AI as a competitive advantage, and 35% are already using it, with nearly half considering adoption soon explodingtopics.com.

🧭 Outlook & Implications

  1. From Tools to Agents: AI is evolving toward systems that autonomously navigate, plan, and execute—requiring fresh thinking about safety and control.

  2. Safety in the Spotlight: Subliminal learning emphasizes hidden risks in synthetic data pipelines, prompting stricter validation.

  3. Global Infrastructure Race: Nations are in a compute arms race, with heavy investment and cross-border collaborations.

  4. Market Explosion: AI’s economic scale is entering uncharted territory—from VC funding to industrial computing needs.

  5. Policy Divergence: While the U.S. deregulates and accelerates deployment aggressively, critics warn about societal and environmental costs.

🔜 What to Watch

  • OpenAI DevDay 2025 (October 6), where more “agentic” capabilities and advanced APIs are expected to be unveiled The VergeLos Angeles TimesOpenAI.

  • UK and U.S. infrastructure buildouts—compute capacity, data centers, HPC partnerships.

  • Ethical balances: carbon-footprint minimization, bias mitigation, transparency in generative ecosystems.

AI today remains a battlefield of economic opportunity, societal promise, and ethical quandary. From autonomous agents to global policy shifts and green computing, we’re in the midst of a transformative moment. Stay sharp and informed—this wave has barely crested.