May 25 Roundup: Gemini Spark launches, OpenAI targets trillion-dollar IPO, Anthropic turns profitable, and AI reshapes banking
Google's I/O keynote unveiled Gemini Spark as a 24/7 personal AI agent with full Workspace integration. OpenAI is racing toward a September IPO that could value the company at $1 trillion. Anthropic is approaching its first-ever profitable quarter with $10.9 billion in Q2 revenue. Major banks including HSBC and Standard Chartered are announcing AI-driven workforce cuts. And the hacking capabilities of frontier models like Anthropic's Mythos are jolting policymakers in Washington — with Trump pulling an AI oversight order at the last minute.
1. Google Gemini Spark: Your AI That Runs While You Sleep
At Google I/O 2026 last week, CEO Sundar Pichai unveiled what he called "the next evolution of smart digital assistants" — Gemini Spark, a 24/7 cloud-based personal AI agent built on Gemini's base models and Google's Antigravity agentic harness. Unlike previous AI assistants that required a user to be actively in the loop, Spark runs on dedicated virtual machines on Google Cloud and continues working even when a device is turned off.
Pichai's pitch was direct: "It's your personal AI agent that helps you navigate your digital life, taking action on your behalf and under your direction. It runs on dedicated virtual machines on Google Cloud seamlessly — you don't need to keep your laptop open to make sure it's running."
Spark ships with deep integration into Gmail, Google Docs, Sheets, and Slides out of the box — a significant distribution advantage over rivals like Anthropic's Claude Cowork and OpenAI's ChatGPT agent, which require users to set up third-party connections manually. Users can email Spark directly through a dedicated Gmail address. On mobile, Android's new Halo system lets you track the agent's progress in real time. Spark also connects to external services over MCP, with more integrations rolling out in the months ahead.
"Need to send an email to your boss with a status update? Spark can pull all the facts from your emails, your docs, your sheets, and slides and write the draft for you. Small businesses are using Spark. They can watch over their inbox, so they never miss a question from a customer." — Josh Woodward, VP, Gemini App & AI Studio, Google
Spark is currently in testing and will roll out to Google AI Ultra subscribers first. The company also announced a new "Daily Brief" agent that organizes the day ahead overnight — gathering information, setting priorities, and suggesting next steps before the user even wakes up. Google shared that Gemini now has 900 million regular users — matching OpenAI's self-reported numbers — and processes over 3.2 quadrillion tokens per month, a 7x jump from just one year ago.
Spark is the most credible ambient AI agent announced by any major lab to date — not because of capability, but because of distribution. Google already owns your inbox, calendar, docs, and browser. Competitors have to convince users to connect APIs; Spark just works. For SMBs and enterprise teams already in the Google Workspace ecosystem, this is going to feel like gaining an employee who never sleeps. The critical question isn't whether Spark will be used — it's whether enterprises will govern how it's used. Businesses running Workspace should start drafting AI usage policies now, before Spark starts taking autonomous actions on behalf of their employees.
2. OpenAI Moves Toward a Trillion-Dollar IPO
OpenAI is preparing to confidentially file an IPO prospectus with the SEC as soon as late May 2026, according to reporting from The Wall Street Journal, Reuters, CNBC, and multiple other outlets. The company — valued at approximately $852 billion in its last private funding round in January 2026 — is working with Goldman Sachs and Morgan Stanley to structure an offering that could come as early as September 2026 and is reportedly being shaped around a valuation of up to $1 trillion.
The IPO will be among the most consequential public offerings in technology history. The S-1 filing will, for the first time, disclose the company's actual financial performance: its revenue trajectory, compute burn rate, training costs, infrastructure spend, and talent expenses. A Fortune analysis noted the filing will "detail exactly how much cash the company is burning on training its models, serving those models on its existing cloud computing infrastructure, building out more data center capacity, and hiring exceptionally pricey AI talent."
"OpenAI is aiming to go public as early as September and is working with Goldman Sachs and Morgan Stanley on a draft IPO prospectus that it plans to file with regulators soon." — Reuters
Sam Altman has reportedly been pushing hard for a September public debut. The confidential filing process — standard for large tech IPOs — allows OpenAI to receive SEC feedback before making its filing public. The company's previous $122 billion fundraising round closed at an $852 billion post-money valuation in early 2026, making a $1 trillion public market valuation a plausible target given typical IPO pricing dynamics.
The OpenAI IPO will be a watershed moment — not just for the company, but for the entire AI sector. For the first time, the market will be able to price AI's most prominent player based on real numbers rather than hype cycles. The filing will force OpenAI to answer questions that have never been publicly answered: How profitable is ChatGPT? What does it actually cost to run a frontier model at scale? Are the economics improving or deteriorating? For enterprise buyers, investors, and competitors alike, the S-1 will be one of the most important AI documents of the decade. Start clearing calendar time for the week it drops.
3. Anthropic Approaches Its First Profitable Quarter — But It May Not Last
Anthropic has told investors that it will more than double revenue to approximately $10.9 billion in Q2 2026 and is on track to deliver an operating profit for the first time in its history, according to reporting from The Wall Street Journal, Bloomberg, and TechCrunch. The milestone is a striking reversal for a company that was burning cash at enormous rates as recently as late 2025.
The revenue surge has been driven by surging enterprise adoption of Claude — particularly among professionals who have expressed a growing preference for Claude over ChatGPT for high-stakes knowledge work. Anthropic has also expanded its customer base aggressively, announcing new services for small businesses and specialized tools for law firms and other professional services verticals.
CNBC named Anthropic the No. 1 company on its 2026 Disruptor 50 list, the first time Anthropic has leapfrogged OpenAI for the top spot. However, the WSJ report noted that profitability may be short-lived: large compute costs are scheduled to hit the company's books later in the year as it scales up infrastructure for the next generation of models.
"Anthropic will more than double revenue to around $10.9 billion in its second quarter, and deliver an operating profit for the first time." — TechCrunch, reporting on Wall Street Journal
Timing matters: the profitability news landed on the same day that OpenAI's IPO plans broke publicly — a striking juxtaposition of two rival narratives. Anthropic's profitability positions it favorably in a potential fundraising context and gives it leverage in enterprise sales cycles where customers want assurance of vendor longevity.
Anthropic's Q2 profitability is real and meaningful — but the caveat about future compute costs is important context that the headlines tend to bury. The company is essentially riding a revenue wave that outpaced its infrastructure spend this quarter, but the next generation of Mythos and post-Mythos models will require massive training runs that will flip that equation. The more durable signal here is enterprise adoption. Anthropic surpassed OpenAI in enterprise AI adoption (34.4% vs. 32.3%) for the first time in recent surveys. That stickiness — driven by trust, safety messaging, and enterprise integrations — is harder to reverse than quarterly profit margins.
4. AI Hacking Models Jolt Washington — And Spark a Delayed Oversight Response
A Politico investigation published May 24 laid out how Anthropic's Mythos model and OpenAI's GPT-5.5 are rattling policymakers and the cybersecurity industry simultaneously — for different reasons. Security researchers testing both models have described their hacking capabilities as a "game-changer," with some claiming Mythos is capable of generating a SolarWinds-scale supply chain attack "every quarter."
Business Insider reported that when Anthropic first disclosed Mythos's capabilities, the company revealed the model had "found thousands of severe security vulnerabilities, some that humans had missed for over a decade," while also being capable of exploiting "every major operating system and every major web browser." Some Anthropic engineers reportedly used the technology to find working exploits over a single night.
"[Some described Mythos as capable of generating] a SolarWinds every quarter, referring to the Russian government's breach of U.S. federal agencies in 2020. The incident is widely regarded as one of the worst hacks in history and affected more than 18,000 organizations worldwide." — Politico
Jonathan Trull, CISO of security firm Qualys (which is testing GPT-5.5), said the OpenAI model "can basically do what your most advanced app security engineer can do." Reuters offered a counterpoint, reporting that while Mythos improves vulnerability discovery, the main challenge remains validating and fixing the flaws identified — a bottleneck that limits the most catastrophic offensive uses.
Against this backdrop, the White House had planned to sign an executive order establishing a voluntary pre-release review process for frontier AI models — potentially requiring a 90-day federal review window before models are released publicly. But President Trump pulled back at the last minute, telling reporters: "I didn't like certain aspects of it. I postponed it." Politico reported that AI czar David Sacks raised industry concerns that contributed to the delay. The directive had explicitly been drafted as voluntary — stipulating it would not create "mandatory governmental licensing, preclearance, or permitting" for AI models.
The Mythos debate is playing out on two tracks simultaneously. On the defensive track, these models represent genuine breakthroughs for security teams — finding vulnerabilities at scale that would take human engineers months to identify. On the offensive track, the same capabilities in the wrong hands are alarming. The White House pullback on voluntary review reflects a real tension in the deregulatory AI posture: the administration wants to avoid slowing innovation, but frontier models are now clearly capable of infrastructure-grade offensive operations. Expect this regulatory standoff to intensify. Organizations running sensitive infrastructure should be treating AI-assisted vulnerability discovery as a near-term threat model, not a future concern.
5. Banks Begin AI-Driven Job Cuts — HSBC and Standard Chartered Lead the Way
Two of Europe's largest banks made headlines this week with unusually frank assessments of AI's impact on their workforces. Standard Chartered announced it would cut approximately 7,800 roles — the first major global bank to explicitly quantify AI-driven headcount reductions. HSBC's CEO Georges Elhedery followed with an investor day speech telling his 200,000 employees to stop resisting the change and embrace it: "We all know generative AI will destroy certain jobs and will create new jobs."
Reuters calculated that HSBC's plan to cut 15% of its corporate function roles by 2030 would result in more than 7,000 redundancies from the more than 52,000 employees in those functions. HSBC has committed to retraining its workforce to help employees become "future ready" — but Elhedery's language was notably unsentimental: "However many will be left at the end of the journey isn't the problem."
"We all know generative AI will destroy certain jobs and will create new jobs. But my initial mission is I need 200,000 colleagues with us on this journey. However many will be left at the end of the journey isn't the problem. The problem is how can we make sure that those 200,000 colleagues have been given all the capabilities, the training, the tools to make themselves future ready." — Georges Elhedery, CEO, HSBC
Standard Chartered's CEO Bill Winters used starker language still, stating the bank wanted to replace "lower-value human capital" with technology. Both banks are operating in highly competitive environments where margin pressure has been acute, and AI's ability to automate compliance checking, report generation, and routine client communication is particularly relevant to the financial sector.
What's happening at HSBC and Standard Chartered is the beginning, not the end, of a broader reckoning across professional services. Banking moves faster than most sectors when cost pressure hits — and the math on AI-driven automation in financial back-office and compliance functions is unambiguous. The more important signal for other industries is the framing: these CEOs aren't calling this a pilot or an experiment. They're publicly quantifying job cuts years in advance and presenting retraining as the mitigation, not the prevention. HR leaders and managers in knowledge-work-heavy industries should take this as a planning signal, not a distant warning.
6. The EU AI Act Hits Full Deployment — Two Years of Reckoning Begin in August
While Washington debates whether to impose even voluntary AI oversight, Europe is approaching a firm deadline. The EU AI Act — which entered into force on August 1, 2024 — becomes fully applicable on August 2, 2026, less than 70 days away. Prohibited AI practices (such as social scoring and certain biometric surveillance systems) have been banned since February 2025. Governance rules and obligations for general-purpose AI models went live in 2025. But the full requirements — including conformity assessments, technical documentation, and transparency obligations for high-risk AI systems — will apply from August 2.
Meanwhile, Colorado became the latest U.S. state to enact its own AI law this week. The revised Colorado AI Act will regulate the use of automated decision-making technology in "consequential decisions" — areas like employment, credit, education, and healthcare — removing the earlier "high-risk AI systems" framing in favor of broader ADMT coverage. The White House's National Legislative Policy Framework for Artificial Intelligence, published in March 2026, had explicitly recommended federal preemption of state AI laws, but that framework remains non-binding as Congress has not yet acted.
"The AI Act will be fully applicable 2 years later on 2 August 2026, with some exceptions: prohibited AI practices and AI literacy obligations entered into application from 2 February 2025." — European Commission
August 2, 2026 is an operational deadline, not a policy aspiration. Any business deploying AI in Europe in high-risk categories — HR decisions, credit underwriting, educational assessment, safety systems — needs to have its conformity assessments, documentation, and transparency mechanisms in place. The EU's enforcement mechanisms are real and the fines are substantial. For U.S. companies operating in European markets, the patchwork of state laws (California, Colorado, Texas, and now others) is creating compliance complexity that mirrors, at a smaller scale, what European firms faced building toward GDPR. The companies that start building structured AI governance programs now will spend less and move faster than those who wait for the enforcement notices to arrive.
7. Model Wars: Gemini 3.5 Flash Is GA, Google Processes 3.2 Quadrillion Tokens a Month
Beyond Gemini Spark, Google's I/O week surfaced a string of model and infrastructure milestones. Gemini 3.5 Flash is now generally available, with Google positioning it as "frontier-level intelligence at 4x the speed of comparable models" — priced at $1.50/$9 per million input/output tokens with a 1-million-token context window. On the developer side, Google launched Managed Agents in the Gemini API in public preview, enabling developers to build and deploy autonomous, stateful agents that run in secure, isolated Google-hosted Linux sandbox environments.
The scale numbers Sundar Pichai shared in his keynote were striking: Google processed 3.2 quadrillion tokens per month across its surfaces — up 7x from a year ago and up from 480 trillion just 12 months prior. Over 8.5 million developers are building with Google's models monthly, and over 375 Google Cloud customers each processed more than 1 trillion tokens in the past 12 months.
On benchmarks, independent evaluations confirm that Claude Mythos outperforms GPT-5.5 on attack chain progression — the specific capability that is generating the most concern in cybersecurity circles. Gemini 3.5 Flash scored 76.2% on Terminal-Bench 2.1, beating Gemini 3.1 Pro on both coding and agent tasks, according to llm-stats.com tracking.
The token volume numbers from Google are the kind of data point that tends to get overlooked in the product announcement noise — and they shouldn't be. 3.2 quadrillion tokens per month represents an extraordinary scale of AI inference, and the 7x year-over-year growth rate implies this is still early adoption. The cost economics of that scale matter enormously for enterprise buyers: as inference costs continue to drop and throughput expands, the business case for AI-augmented workflows improves dramatically. Gemini 3.5 Flash at $1.50/million input tokens is now competing on cost with open-source inference setups — which will accelerate adoption and make the "build vs. buy" calculation much simpler for most organizations.
⚡ Why This Week Matters
This week's headlines converge on a single theme: AI is moving from pilots and demos into operating infrastructure — with real financial stakes, real workforce consequences, and real regulatory deadlines attached.
- 🤖 Gemini Spark makes always-on AI agents a mainstream product for the first time — and Google's Workspace integration gives it the distribution moat that rivals lack.
- 📈 OpenAI's IPO filing will be a transparency moment for the entire industry — the first real look at the unit economics of frontier AI at scale.
- 💰 Anthropic's profitability is a milestone, but the compute cost wall ahead means this is a sprint to build enterprise stickiness, not a comfortable cruise.
- 🔐 AI cyber capabilities are now genuinely alarming policymakers — and the Trump administration's indecision on oversight creates a vacuum that attackers could exploit.
- 👔 Banking job cuts are the canary in the coal mine for professional services AI displacement — the timeline is accelerating and the scale is now being stated in public.
- ⚖️ August 2 is the EU AI Act's full compliance deadline. If your business touches European markets with AI systems, the clock is ticking.
Need help navigating AI for your business?
From Gemini Spark adoption strategies to EU AI Act compliance, our team turns these developments into actionable strategy for your organization.
Contact SEN-X →