If Week 1 was about the hardware race and the corporate restructuring happening around AI, Week 2 was about something bigger: institutions. A court delivered a verdict on the most watched legal dispute in AI. A religious institution published its first formal position on artificial intelligence and human dignity. A professional services giant deployed AI to a quarter of a million employees simultaneously. And a government cancelled an executive order on AI, citing competitive strategy rather than safety.
The
theme across all of it is not technical. It is structural. The decisions being
made this week about who governs AI, who is accountable for it, and who has
access to it will shape the next decade of the technology's development more
than any model release.
STORY 1
Google I/O
Delivered More Than Expected. Here Is What Actually Matters.
Google's annual developer conference on May 19 produced a volume of announcements that is genuinely difficult to process in a single sitting. The AI Vanguard is covering the most consequential developments.
The headline model release is Gemini 3.5 Flash, which Google made available immediately on the day of the announcement and has already deployed to power Google Search and its Antigravity 2.0 product. Gemini Omni, a unified model processing text, images, and video natively without transcription intermediaries, is the more technically significant announcement. Google is also rebuilding Search itself, introducing a feature that generates custom visuals, interactive graphics, and mini-applications directly within search results rather than simply surfacing links.
The
integrations announcement is the one most relevant to everyday users. Adobe,
Canva, and CapCut announced Gemini integrations that allow users to generate AI
content in Gemini and then refine it in professional editing tools without
switching applications. Canva is live for Gemini AI Ultra subscribers. Adobe
and CapCut are rolling out in coming weeks.
The AI Vanguard Take:
Google I/O
confirmed that Google is not trying to win the AI chatbot race. It is trying to
make AI the infrastructure layer underneath every Google product and every
creative tool that integrates with Google. That is a fundamentally different
competitive strategy from building the best standalone chatbot, and it is one
that plays to Google's existing strengths more directly than anything it has
attempted in AI so far.
STORY 2
Musk Lost His
OpenAI Lawsuit. The Jury Deliberated for Less Than Two Hours.
A California federal jury on May 19 unanimously rejected all of Elon Musk's claims against OpenAI and its chief executive Sam Altman. The jury deliberated for less than two hours after eleven days of testimony. The ruling: all of Musk's claims were barred by the statute of limitations. He had waited too long to file.
Musk co-founded OpenAI in 2015, departed in 2018, and subsequently argued in his lawsuit that the company had violated its nonprofit founding mission by converting to a for-profit structure. OpenAI countered that there was never a binding commitment to remain nonprofit and characterised the lawsuit as a tactical attempt to slow a competitor.
The
verdict removes a significant legal cloud from OpenAI's path to a public market
listing. It also closes, at least in this legal forum, the question of whether
OpenAI's transition from nonprofit to for-profit constituted a breach of its
founding commitments. That underlying question of institutional accountability
does not disappear with the verdict, but it will now need to be addressed
through policy and governance rather than litigation.
The AI Vanguard Take:
The speed of
the jury's deliberation was itself a statement. Eleven days of testimony and
less than two hours of deliberation suggests the legal team brought a weaker
case than the public controversy surrounding it implied. Whether the founding
mission argument has merit as a matter of ethics is a separate question from
whether it had merit as a legal claim, and the verdict answers only the second
of those.
STORY 3
KPMG Deployed
Claude to 276,000 Employees Across 138 Countries.
On May 19, the same day as Google I/O and the Musk verdict, KPMG and Anthropic announced the KPMG Digital Gateway Powered by Claude. The deployment covers 276,000 employees across 138 countries, making it one of the largest single enterprise AI deployments ever announced.
The significance of this announcement extends beyond the scale. KPMG is one of the four largest professional services firms in the world, serving clients in audit, tax, and advisory across virtually every major economy. Deploying Claude at this scale to a workforce handling sensitive financial, legal, and strategic client work signals that Anthropic's enterprise data handling commitments are considered sufficient by one of the world's most risk-conscious professional service organisations.
This
announcement landed alongside news that OpenAI launched a consulting subsidiary
called DeployCo with 150 forward-deployed engineers, directly copying the
embedded engineer model used by Palantir. The two announcements together signal
that the most important competitive battle in enterprise AI is no longer about
which model performs best on benchmarks. It is about which company can deploy
AI most effectively inside the organisations that are paying for it.
The AI Vanguard Take:
KPMG choosing
Claude over ChatGPT for a deployment of this scale is a significant vote of
confidence in Anthropic's enterprise positioning. The race for enterprise AI
deployment is increasingly about trust, integration capability, and
professional service support rather than raw model capability. Both Anthropic
and OpenAI are structuring accordingly.
STORY 4
The Pope
Published an Encyclical on Artificial Intelligence.
On May 25, Pope Leo XIV published Magnifica Humanitas, described as the first papal encyclical specifically addressing artificial intelligence and human dignity. It was signed on the 135th anniversary of Rerum Novarum, Pope Leo XIII's landmark 1891 encyclical on the rights of workers in the industrial age. The deliberate symbolism of the date is its own argument: the Church is positioning AI as a civilisational challenge comparable to industrialisation.
The encyclical addresses AI's impact on human relationships, creative work, labour markets, concentration of power, and autonomous weapons. It does not reject AI but argues that technology must serve human dignity rather than diminish it, and calls for international governance frameworks that distribute AI's benefits broadly rather than concentrating them among those who control the infrastructure.
In a
detail that has received less coverage than it deserves, Anthropic co-founder
Christopher Olah presented the encyclical alongside cardinals at the Vatican
Synod. The presence of a senior AI researcher at the launch of a papal document
on AI reflects the seriousness with which both the Church and the AI safety
community are engaging with each other on questions of governance and ethics.
The AI Vanguard Take:
The Catholic
Church has 1.4 billion members. When it publishes a formal doctrinal position
on AI, it is not a marginal cultural moment. Magnifica Humanitas will shape how
hundreds of millions of people think about AI ethics, particularly in Latin
America and sub-Saharan Africa where Catholicism is growing rapidly. The
framing of AI as a justice and labour rights question, echoing Rerum Novarum,
is more politically consequential than most technology policy documents
published this decade.
STORY 5
Chinese
Open-Source Models Now Account for 60 Percent of AI Usage on OpenRouter.
OpenRouter, the most widely used third-party AI model router that allows developers to access multiple AI models through a single API, published usage statistics this week showing that Chinese models now account for 60 percent of all usage on the platform. The models leading this share are Kimi K2.6, DeepSeek V4, GLM-5.1, and Qwen 3.
This is the clearest single data point yet that the open-weights tier of the AI model market, the segment where models are freely available for download and deployment, is now effectively Chinese-led. Meta's Llama models, which once dominated open-source AI usage, have been displaced. Meta's next-generation frontier open-weights model, internally named Avocado, has been delayed and has gone quiet with no announcement.
For
developers, researchers, and businesses outside China who have built their
workflows on open-source AI, this is a strategic consideration rather than just
a market share statistic. Questions about data handling, training data
provenance, and alignment approaches in models from Chinese labs are legitimate
considerations that do not yet have fully transparent answers.
The AI Vanguard Take:
The
open-source AI race is being won by Chinese labs with a speed and margin that
Western commentators have been slow to acknowledge. This matters not because
Chinese-made AI is inherently untrustworthy, but because the governance,
transparency, and accountability frameworks that apply to these models differ
significantly from those applied to US-developed alternatives. That difference
deserves honest discussion rather than either dismissal or panic.
STORY 6
Anthropic Is
Closing a $30 Billion Round at a $900 Billion Valuation.
Bloomberg reported this week that Anthropic is finalising terms on a $30 billion fundraising round at a valuation of $900 billion or higher, co-led by Sequoia, Dragoneer, Greenoaks, and Altimeter. If the round closes at the upper end, Anthropic surpasses OpenAI's March 2026 valuation of $852 billion, representing a complete reversal from February when Anthropic was valued at $380 billion.
The financial context makes the valuation less surprising than it sounds. Anthropic shared projections with investors showing $10.9 billion in revenue for Q2 2026, up 130 percent from $4.8 billion in Q1. Claude powers Cursor and Windsurf, the two most widely used AI coding editors. The KPMG deployment announced this week adds further institutional revenue. The company is growing faster than almost any software business in history.
The
AI Vanguard covered Anthropic's $900 billion valuation in the Week 1 digest but
the confirmation this week from Bloomberg with investor names and revenue
projections attached moves it from report to near-certainty. The SpaceX IPO
scheduled for June 12 will be the next major data point for how the market
values AI infrastructure companies at scale.
The Week's Theme in One Sentence
The
most important AI stories of Week 2 were not about what AI can do. They were
about who governs it, who is accountable for it, and who gets to use it:
questions that courts, religious institutions, professional firms, and
governments are now answering simultaneously, and not always in compatible
ways.
What to Read on The AI Vanguard This Week
•
How to Use Claude AI Like a Pro
•
The Student's Guide
to AI and the Cerebras IPO analysis
•
How to Automate
Your Work with AI and Can AI Write Better Than Humans?
•
AI Laws and Regulation and the 10 Best AI Productivity Tools
•
Deepfakes AreGetting Scary Good and How Solo Entrepreneurs Are Using AI
•
How to Use Gemini AILike a Pro and the Best AI Tools for Content Creators
• AI in Healthcare
and How to Use AI for Marketing
