June 2 Roundup: Anthropic Files for IPO, GPT-5.5 Instant Goes Default, Gemini Omni Lands, Mythos Hits EU, and CNN Sues Perplexity
The AI industry hits a new inflection point: the two biggest labs are racing to Wall Street, models are getting sharper faster than most businesses can adapt, and the content industry is fighting back hard. Here's everything you need to know from the past 24 hours.
1. Anthropic Confidentially Files for IPO — Valuation Approaches $1 Trillion
In one of the most anticipated market events of 2026, Anthropic announced Monday that it has confidentially submitted a draft registration statement (Form S-1) with the U.S. Securities and Exchange Commission for a proposed initial public offering. The move positions the Claude maker — now valued at approximately $965 billion — as the frontrunner in a closely watched race to bring major AI labs to public markets.
The filing comes just days after Anthropic closed a massive $65 billion Series H round led by Altimeter Capital, Dragoneer, Greenoaks, Sequoia Capital, Capital Group, Coatue, and D1 Capital Partners. That round pushed its post-money valuation to $965 billion, surpassing rival OpenAI's $852 billion post-money valuation from its own March fundraise. Anthropic has not yet set the number of shares or the offering price, and the company noted the IPO will depend on market conditions and other factors.
"Anthropic PBC has confidentially filed draft paperwork to go public as it races rival OpenAI to make a Wall Street debut as soon as this fall. The number of shares to be offered and the price have not been set yet."
— Bloomberg, June 1, 2026
Anthropic's revenue trajectory makes the filing credible. The company recently disclosed a revenue run-rate surpassing $47 billion — up from $9 billion at the end of 2025. That rapid growth is driven largely by enterprise adoption of Claude and growing capabilities like the restricted Mythos model (more on that below). The company was founded in 2021 by former OpenAI employees Dario and Daniela Amodei, and was for years considered a distant competitor to OpenAI's ChatGPT. That positioning has flipped.
OpenAI, meanwhile, is expected to file its own IPO prospectus in the coming days or weeks, setting up an historic parallel AI lab debut that will test institutional investor appetite for the sector at scale. SpaceX, which is also targeting a $2 trillion valuation IPO and seeking to raise more than $75 billion, adds to the density of the 2026 IPO calendar.
The Anthropic IPO filing is the clearest signal yet that the first phase of AI development — the "fundraise and build" era — is giving way to a public accountability phase. When Anthropic files its full S-1, the world will get its first detailed look at the economics of running a frontier AI lab: compute costs, customer concentration, competitive moats, and margin structure. That transparency will reshape how enterprises think about vendor lock-in, long-term pricing, and the stability of their AI supply chain. For business leaders evaluating AI strategy today, the IPO season is a research opportunity as much as a financial event. Watch the S-1 disclosures carefully.
2. OpenAI Makes GPT-5.5 Instant the Default — GPT-4.5 and o3 Get Sunset Dates
OpenAI has confirmed that GPT-5.5 Instant is now the default model in ChatGPT, replacing the previous rotating defaults as the company continues its rapid model refresh cadence. The update brings cleaner, more natural replies alongside in-chat writing and coding blocks — features that reduce friction for everyday knowledge-worker use cases.
Alongside the GPT-5.5 Instant update, OpenAI confirmed two significant model retirements. GPT-4.5 will be retired from ChatGPT on June 27, 2026, following a 30-day sunset period. OpenAI o3 — which powered many of the company's most impressive reasoning demonstrations — will be retired on August 26, 2026, following a 90-day sunset window. Both retirements reflect OpenAI's accelerating belief that newer reasoning-integrated models make the older specialized architecture less necessary.
"OpenAI o3 will be retired from ChatGPT on August 26, 2026 following a 90-day sunset period, and GPT-4.5 will be retired from ChatGPT on June 27, 2026 following a 30-day sunset period."
— OpenAI Help Center, June 2026
Looking further ahead, GPT-5.6 is expected to ship in June 2026 with notable advances in agentic workflows and advanced reasoning. Reports suggest GPT-5.6 will introduce a dedicated "Pro" variant with enhanced multi-step task execution capabilities. OpenAI's rapid versioning pace — from 5.3 to 5.5 in the space of weeks — reflects the competitive pressure from Anthropic's surging Claude product line and Google's Gemini push.
For enterprise customers, the practical implications are real: workflows built on specific model versions need ongoing auditing, and the value of vendor-agnostic prompt engineering keeps increasing. API consumers have slightly more runway with model deprecation timelines announced in advance, but the message is clear — OpenAI intends to keep turning the model stack over quickly.
OpenAI's model retirement cadence is becoming a business continuity issue for enterprises that haven't invested in abstraction layers. GPT-4.5 retiring in under a month is a sharp wake-up call. If your workflows are tightly bound to specific model identifiers — particularly for structured output formats, safety behaviors, or edge-case handling — you need a testing and migration protocol today, not after the sunset date. The opportunity here is real though: GPT-5.5 Instant's cleaner outputs and better coding blocks genuinely improve productivity for the majority of enterprise use cases. Move to it intentionally rather than by default.
3. Google I/O Week's Biggest Launches Begin Rolling Out: Gemini Omni and 3.5 Flash Go Live
Google's blockbuster Google I/O 2026 announcements from last week are now hitting users' hands. The two standout launches — Gemini 3.5 Flash and Gemini Omni — represent a significant shift in Google's AI product strategy: away from model benchmarking braggadocio and toward agentic, multi-step practical utility.
Gemini 3.5 Flash is now generally available via Google Antigravity, the Gemini API, and Google AI Studio. According to Google, it delivers frontier-level intelligence comparable to large flagship models at Flash-series speed — outperforming Gemini 3.1 Pro on the Terminal-Bench 2.1 coding benchmark (76.2%), GDPval-AA agentic benchmark (1656 Elo), and MCP Atlas (83.6%). It's designed for long-horizon agentic tasks and is particularly effective at rapid application development, codebase maintenance, and financial document preparation. Google says tasks that previously took developers days can now be completed in hours or less, at under half the cost of competing frontier models.
Gemini Omni is Google's new multimodal creation model — described as capable of generating "anything from any input, starting with video." It combines Gemini's reasoning capabilities with generative media models, incorporating physics simulation for more realistic video outputs including gravity, kinetic energy, and fluid dynamics. All content generated by Gemini Omni includes SynthID digital watermarks for AI provenance tracking.
"Our biggest, boldest new developments took center stage at Google I/O 2026. We announced technical breakthroughs, like Gemini Omni's ability to create anything from any input, starting with video."
— Google Blog, May 2026
Sundar Pichai declared the "agentic Gemini era" at I/O, announcing Gemini 3.5 Flash availability, the Gemini Spark personal agent, and a pricing restructure that drops Gemini AI Ultra from $250 to $200 per month. Gemini now counts 900 million monthly active users in the app — a number that puts it within reach of ChatGPT's scale for the first time. Gemini 3.5 Pro, already in internal use, is expected to roll out next month.
Google's I/O push matters more than the benchmarks suggest. The combination of Gemini 3.5 Flash's sub-frontier-price performance with Gemini Omni's multimodal generation capability gives enterprises a genuinely capable, deeply integrated stack — one that connects document intelligence, video creation, web search, and now agentic task execution in a single provider relationship. For businesses already in Google Workspace, the friction to adopt is extremely low. The 900M user figure also signals that Gemini has become the default AI touchpoint for a large portion of Google's user base. That distribution advantage is hard to replicate.
4. OpenAI Launches Rosalind Biodefense Program — AI Enters Pandemic Preparedness
OpenAI has formally launched the Rosalind Biodefense Program, a new initiative that offers its specialized GPT-Rosalind life-sciences model to trusted developers and U.S. federal agencies for the purpose of building biodefense and pandemic preparedness tools. The program, named for pioneering chemist Rosalind Franklin, marks OpenAI's most direct foray into dual-use life sciences AI to date.
GPT-Rosalind is a biology-focused model designed for biosurveillance, disease monitoring, biodefense research, and public health threat assessment. OpenAI is offering early access to vetted developers and government agencies, with the explicit goal of "strengthening societal resilience" before the next biological threat emerges. The company says the program will help operationalize biodefense tools — moving AI capabilities out of research labs and into functional monitoring and response infrastructure.
"This program helps trusted developers apply frontier AI capabilities to operationalized biodefense tools that can strengthen preparedness before the next biological threat emerges."
— OpenAI, June 2026
The launch has drawn attention beyond the AI industry because it sits squarely at the intersection of AI's most significant dual-use risks. Biology-focused AI with high capability represents both an enormous public health opportunity and a potential misuse vector. OpenAI's response is a trust-gating model: restricting GPT-Rosalind access to screened organizations, while making the capability available to those who need it most for legitimate defense work.
The Rosalind program complements Anthropic's parallel Project Glasswing effort (focused on cyber vulnerability detection) and signals that frontier AI labs are increasingly positioning their most powerful models as public infrastructure tools — with carefully managed access tiers rather than open availability.
The Rosalind Biodefense launch is a strategic signal that matters regardless of whether your business has any connection to life sciences. It demonstrates that frontier AI labs are now serious enough — and powerful enough — to position themselves as partners in national security and public health infrastructure. That positioning will accelerate government AI spending, deepen regulatory entanglements, and raise the bar on what "responsible AI deployment" means in practice. For enterprise customers, particularly in healthcare, pharma, and public sector, Rosalind also previews what specialized vertical models will look like: not general models with safety layers bolted on, but purpose-trained, access-controlled tools built for high-stakes domain work.
5. Anthropic Offers EU Access to Mythos — Europe's AI Sovereignty Bet Gets Real
Anthropic has informed the European Union that it is willing to provide EU governments and institutions with access to its powerful but still-restricted Mythos model — a development that signals a significant shift in the transatlantic AI relationship. CNBC first reported the overture, which comes as the EU has been in direct talks with OpenAI, Anthropic, and other frontier labs about access to advanced AI capabilities as part of its broader AI sovereignty strategy.
Mythos, previewed in April 2026 as part of Anthropic's Project Glasswing cybersecurity initiative, has already demonstrated extraordinary capabilities: it identified more than 10,000 previously unknown zero-day vulnerabilities across major operating systems and web browsers during Glasswing's initial rollout phase. That level of capability — essentially an AI-powered penetration testing and vulnerability discovery engine — makes Mythos both a defense windfall and a governance challenge.
"We welcome the latest developments on potential future access. The bloc aims to get a clearer idea of the potential risks that the technology poses."
— EU tech sovereignty spokesperson Thomas Regnier, via CNBC
The EU's cautious welcome — it wants to understand risks before committing — reflects the bloc's ongoing tension between AI competitiveness and precautionary governance. Europe has been reluctant to be left behind in frontier AI access while also insisting on rigorous safety review. Anthropic's willingness to engage on EU Mythos access may also be a smart pre-IPO move: demonstrating responsible international deployment of powerful capabilities ahead of public market scrutiny.
For wider Mythos rollout, Anthropic has indicated the model will become more broadly available as the company works through its restricted access program. Forbes reported that partners have already used Mythos Preview for weeks to conduct local vulnerability detection, black-box testing, endpoint security assessments, and penetration testing — and results have been strong enough that demand for broader access is building fast.
Mythos is the sharpest illustration yet of a pattern we've been tracking: the most powerful AI models are not being released broadly — they're being channeled through institutional gatekeepers. Anthropic controls who gets Mythos access. OpenAI controls who gets GPT-Rosalind access. This trust-gating model is sensible for dual-use risks, but it also creates a new kind of AI haves and have-nots at the enterprise level. Organizations that build the security vetting relationships with labs now — through programs like Glasswing, government partnerships, or enterprise research agreements — will have access to capabilities that others simply can't buy. That access gap will matter enormously within 18 months.
6. CNN Sues Perplexity for Copying 17,000 Stories — And State AI Laws Keep Multiplying
The AI industry's legal and regulatory environment shifted sharply this week on two fronts. CNN has filed suit against Perplexity AI in the U.S. District Court for the Southern District of New York, accusing the AI search startup of unlawfully copying and distributing more than 17,000 CNN stories, videos, images, and other published works to power its products without authorization or compensation.
The lawsuit alleges that Perplexity "unlawfully crawls, scrapes, copies, and distributes" CNN's content — a more aggressive framing than many prior AI copyright suits that focused on training data. Perplexity's defense is expected to lean on arguments that facts cannot be copyrighted and that its use constitutes fair use. The case joins a rapidly expanding docket of AI copyright litigation that includes suits from major publishers, music labels, and creative professionals against OpenAI, Anthropic, Google, and others.
"CNN's lawsuit alleges that Perplexity violated federal copyright law by copying more than 17,000 CNN stories, videos, images, and other content to power its products and tools."
— The Statesman, May 2026
On the regulatory front, states continue to fill the federal vacuum. Colorado enacted Senate Bill 26-189 (the ADMT Act) in May, repealing and replacing its earlier Colorado AI Act with a more limited approach to AI regulation effective January 1, 2027. Vermont's legislature passed a consumer data privacy bill. Illinois' legislature passed an AI frontier model bill, requiring safety evaluations for models above certain capability thresholds. And at least eight AI-related bills crossed state legislative milestones in the past week alone. Courts are also signaling impatience: NPR reported that federal courts are increasingly overwhelmed with AI-related filings as use cases skyrocket and legal wins fail to keep pace with expectations.
The regulatory picture is heading toward a patchwork of state laws with varying compliance requirements — exactly the outcome the AI industry lobbied against. With Congress still unable to pass federal AI legislation and the White House focused on preemption rather than comprehensive rules, enterprises are now navigating a genuinely fragmented compliance landscape.
The CNN v. Perplexity case is the one to watch in AI copyright law. Unlike earlier training-data suits, this case focuses on operational content use — the ongoing, repeated copying of published works to power live AI products. If CNN prevails on this theory, the implications extend far beyond Perplexity to any AI product that surfaces summaries, answers, or outputs that draw substantially on third-party published content. For enterprises building AI-powered content surfaces, the question is no longer just "did we license our training data?" but "do we have defensible rights to the outputs our AI is generating from live sources?" That is a materially harder question. Pair it with the state-law fragmentation, and the compliance function for AI products is about to get significantly more complex.
🔎 Why It All Matters — The June 2 Picture
Today's stories aren't independent data points — they're a coherent narrative about AI maturing into an industry with real financial accountability (IPO filings), product differentiation through specialized models (Mythos, Rosalind), capability acceleration (Gemini Omni, GPT-5.5 Instant), and growing legal and regulatory friction (CNN suit, state laws). Businesses that treat AI as a technology experiment are increasingly finding themselves managing a strategic business risk. The companies that will win in this environment are the ones that have moved from "AI pilots" to "AI operations" — with governance frameworks, vendor strategies, and legal risk assessments already in place.
The IPO race between Anthropic and OpenAI will produce the most detailed public disclosure of frontier AI economics in history. For anyone making long-term technology bets, reading those S-1s carefully — when they drop — will be among the highest-return research investments of the year.
Need help navigating AI for your business?
Our team turns these developments into actionable strategy — from model selection and vendor risk to governance frameworks and custom AI deployment.
Contact SEN-X →