There is a version of this piece about grief. About strangers who never met Ozzy Osbourne deciding, with considerable conviction, that their emotional connection to his music gives them a meaningful say in what happens to his legacy. It does not. Grief is not ownership. The fans are not shareholders; the grief a Swindon Town supporter feels for the slow death by a thousand cuts each season does not give them a seat in the boardroom either. Interesting questions do exist here, possibly too many before we all reach for the comfort blankets of the bong and the sitar. But one view I do hold is that moral outrage from people whose lives were touched by someone else's art is not a controlling interest.

Anyway. Sombre and reflective is not really my natural register, and this week's article/rant can be more accurately summarized as follows:

Ozzy Osbourne, Dario Amodei and Elon Musk walk into a bar.

Jeff Bezos and Sundar Pichai are in the corner playing pool.

Sam Altman says.

"What will it be?"

The rest, as they say, is financial history.

AI & UK GDPR: What Leaders Get Wrong About Data Protection and AI Risk
Is your AI deployment actually a GDPR problem — or are you solving the wrong issue? Most leaders either over-panic about AI and data protection, or ignore it entirely. Craig Clark, Director of Clark and Company Information Governance Services and a practising CISO, cuts through the noise: AI only becomes a UK GDPR issue when personal data is involved — but when it is, the risks are serious and often misunderstood. Speakers: Craig Clark, Director at Clark and Company Information Governance Services; Stewart Tinson, Projecgt Director, AI-360. You’ll learn: - Why AI is not automatically a GDPR issue — and the one question that determines whether it is - Why consent is not the most important lawful basis, and which lawful basis actually fits AI use cases - Why a global firm’s AI agent hallucinated its entire sales strategy throughout 2026 — and what human oversight would have prevented it - How to approach DPIAs as a design tool, not a compliance tick-box - Why facial recognition bias disproportionately harms people who are not middle-aged white men - What to actually ask AI vendors regarding data usage and sharing Key topics: UK GDPR & AI • Lawful basis for AI • Special category data • DPIAs • Transparency obligations • Human in the loop • AI supplier due diligence • Facial recognition bias • Data minimisation • Security non-negotiables Essential viewing for CISOs, CIOs, legal, and compliance leaders deploying AI in data-sensitive environments.

The Trillion-Dollar Starting Gun

On 3 June, SpaceX filed its amended S-1 with the SEC, pricing its shares at $135 under the ticker SPCX, with net proceeds before expenses expected to reach approximately $74.4 billion. The company has valued itself at $1.75 trillion, a figure that would put Musk's personal holding above $600 billion and make him the world's first trillionaire.

The filing covers a combined entity incorporating xAI, Grok, X, and the COLOSSUS data center infrastructure, following SpaceX's acquisition of xAI in February 2026. Starlink remains the commercial engine: $3.26 billion in Q1 2026 revenue, $1.19 billion in operating income, 10.3 million subscribers in 164 markets. The AI segment tells a different story. Revenue of $818 million in Q1 against a $2.47 billion operating loss, with capital expenditure in that segment alone hitting $7.72 billion in the quarter. The full-year 2025 net loss was $4.9 billion. Legal exposure adds further colour, with over half a billion dollars flagged in anticipated costs covering deepfake lawsuits, patent claims, EU content moderation failures, and data breaches.

None of that will necessarily deter investors. The SpaceX listing bundles rockets, satellites, and a combined AI and social media operation into one offering, and Musk retains approximately 82.4% of voting power post-IPO. Buying in means buying him. That bet has paid out before.

Agents Aren’t Magic: Moving from Excessive Permissions to Governed Workflows
Most organizations are still in what one practitioner calls the “proof-of-concept hangover phase” — giving agents excessive permissions and broad data access just to make the demo work. Wiz’s AI posture research confirms it: agents deployed without tenant isolation, with excessive or potentially admin-level permissions, and unconstrained data access. The question is no longer whether agentic AI is transformative. It’s whether your architecture can actually govern it. Speakers: Aron Eidelman, Security Advocate at Google Cloud, and David Pierce, AI Security and Data Observability Architect at PayPal You’ll learn: • Why the governance response to an agent hallucinating a destructive command and an agent being actively exploited via prompt injection is architecturally the same — and where deception-based detection creates a meaningful exception • How to apply SRE error budgets to agentic AI: when an agent burns through policy violations or triggers guardrails too frequently, autonomous execution degrades gracefully back to human oversight • Why human-in-the-loop approval becomes meaningless the moment alert fatigue sets in — and which decision types still require a human by design • What AI Passports and Google Cloud’s emerging agent identity principal type mean for least privilege and orphaned credential risk • How to quantify security risk in financial terms that justify board-level investment Key topics: MCP attack surface • Ephemeral agents • GFCI control model • Agent error budgets • Failure domain isolation • Friction logs • AI Passports • Agent identity • Deception detection • Risk quantification For CISOs, security architects, and engineering leaders deploying agents into production.

Anthropic and the Art of Being Strategically Vague

Anthropic has filed confidentially. Most recently valued at $965 billion, it has submitted a draft S-1 without disclosing share counts or price ranges, with a listing expected to target $1 trillion or higher and potentially arrive as early as autumn 2026.

The valuation trajectory tells its own story: from $61.5 billion in March 2025 to $965 billion in just over a year, driven by Claude Code, broad chatbot adoption, and a safety reputation that has made it the enterprise AI partner of choice.

For Amazon and Google, this is one of the most significant paper-to-cash conversion events in corporate history. Amazon's position spans convertible notes and nonvoting preferred stock worth approximately $74.2 billion on paper, with $16.8 billion in pre-tax gains booked from that position in Q1 2026 alone and a commitment to invest up to $20 billion more.

Court documents confirm Google holds roughly 14% in straight equity, worth approximately $135 billion at current valuations, with a further $40 billion commitment made in April. Microsoft, which committed $5 billion to Anthropic in late 2025, is also positioned to benefit, albeit from a rather smaller seat at the table.

The relationship between Anthropic and its largest backers runs deeper than equity. Anthropic has committed to spending more than $100 billion on Amazon's infrastructure over the next decade, while AWS sells Claude to enterprise customers via its Bedrock platform. They are not merely investor and investee; they are each other's infrastructure.

Retroactive AI Governance: Why Bolt-On Security Is Setting Enterprises Up to Fail
Many organisations rushed to deploy AI. Now they’re scrambling to bolt on governance after the fact — and discovering they don’t even know what’s in their environment. How do you secure what you can’t see? Celina Stewart, Director of Cyber and AI Risk Management at Neuvik, breaks down why retroactive AI governance isn’t working, where the real risks hide in AI’s unique tech stack, and what executives need to do now before agentic AI makes the problem significantly harder. Speakers: Celina Stewart, Director of Cyber and AI Risk Management at Neuvik, and Stewart Tinson, Project Director, AI-360 Online You’ll learn: • Why companies that implemented AI first are now scrambling to retrofit governance they weren’t built for • How third and fourth-party risk in AI supply chains introduces threats most vendor questionnaires never reach • Why shadow AI and BYOD are creating visibility gaps that make it nearly impossible to implement effective controls • What a third-party risk management programme tailored specifically for AI should actually address — including memory injection controls • How agentic AI demands a shift from zero trust to continuous trust, and what “Know Your Agent” looks like in practice Key topics: Retroactive AI governance • Shadow AI visibility • Third and fourth-party AI risk • AI tech stack security • Insider risk from AI tools • Vendor rebranding and marketplace noise • Agentic AI continuous trust • Know Your Agent (KYA) • AI-specific vendor diligence • Employee AI education Essential viewing for CISOs, CIOs, and enterprise leaders responsible for AI security, governance, and risk management.

The View from OpenAI

Watching two of the most anticipated IPOs in market history land in the same week, one from Musk whose legal assault on OpenAI was dismissed by a jury just days ago, and one from Amodei whose company was co-founded by former OpenAI alumni, Sam Altman's position is not enviable. OpenAI remains privately held. It is just not the one commanding the headlines.

There is one further detail Altman might find particularly galling. When Pope Leo XIV launched his AI encyclical in May, Anthropic co-founder Christopher Olah was seated beside him at the Vatican. For every person who saw the image without reading all 245 paragraphs.....

If both listings achieve their target valuations, they will together represent somewhere between $2.75 trillion and $3.5 trillion in market capitalization arriving on public markets within months of each other. That is not a market event; it is a gravitational field. Bezos and Pichai will be fine, their pool game interrupted only by notifications confirming their Anthropic stake has gone up again. As for Mark Zuckerberg, he does* not appear anywhere in this story. Meta, it seems, decided to build its own bar.

The Ozzy avatar will eventually prompt the questions I originally planned to write about, just not this week. This week, the money spoke louder. It usually does.


AI Literacy: Why Role-Based Training Beats Every Generic Program
Is your organisation’s AI literacy program already out of date before it launched? With AI tools proliferating week by week, generic training programs and 25-page policies are creating a false sense of security — not capability. The gap between informal AI experimentation and structured organisational readiness is where shadow AI estates form, data governance breaks down, and staff learn to click through compliance training without retaining anything. Speakers: Craig Clark, Director at Clark and Company Information Governance Services, and Stewart Tinson, Project Director at AI-360. You’ll learn: Why role-based AI literacy — C-suite, operational, technical, and governance — is the only model that scales How to move from unstructured experimentation to assessed, accountable AI adoption Why measuring effectiveness by attendance alone is a mistake, and what to measure instead How to prevent shadow AI estates and data governance blind spots Why leadership must visibly model responsible AI use or risk programs failing entirely Key topics: Role-based AI capability • Shadow AI estate risk • Enterprise vs. personal tools • Critical thinking and over-reliance • Measuring program effectiveness • Keeping literacy programs current • Leadership modelling • AI as augmentation For enterprise leaders — CISOs, CIOs, and those responsible for AI governance and workforce capability — who need practical frameworks, not theory.

*or did not until I mentioned him.

Share this post
The link has been copied!