Also published as The AI Edge on LinkedIn.
Covering: OpenAI and Anthropic's IPO race · The safety team that vanished before the listing · OpenAI builds on Microsoft's lawn · Europe staffs up for the scrutiny to come
This Week at a Glance
- Anthropic filed a confidential S-1 this week, targeting a ~$2 trillion valuation — roughly twice what it was worth at the start of 2026 and comfortably ahead of Goldman Sachs, which took 131 years to reach the same market cap.
- OpenAI disbanded its Preparedness team — the internal function responsible for catastrophic risk evaluation — eighteen months after creating it, two months before its own expected IPO, and without a named replacement function or lead.
- OpenAI is buying NextSlide, the AI-native presentation company and the most direct challenger to PowerPoint's enterprise stranglehold.
- Anthropic is building out its EU headcount quietly, with new roles in Brussels and Dublin ahead of what looks like a busy regulatory autumn.
Section 1: The Big Story
The Race to List — and What Gets Left Behind
Anthropic filed a confidential S-1 with the SEC this week. The target valuation is approximately $2 trillion, making this one of the largest IPOs in US history if it completes as planned. The current forecast window for pricing and debut is approximately 26 October, give or take a few weeks either side depending on market conditions. The legal structure is worth noting: Anthropic operates under a Public Benefit Corporation charter and a Long-Term Benefit Trust governance layer, which means the IPO itself requires fiduciary reconciliation between shareholder return obligations and the "responsible development and maintenance of advanced AI for the long-term benefit of humanity" clause embedded in the charter. Whether that tension is real or ceremonial will be tested the first time a significant commercial decision conflicts with a stated safety commitment in a way investors can quantify.
OpenAI's situation is different in structure but similar in timing. The company converted from its nonprofit model to a Delaware public benefit corporation in March 2026, resolving the legal uncertainty that had hung over the for-profit subsidiary since late 2024. The current forecast is a 2027 IPO, but the operational decisions being made now are clearly IPO-adjacent: cost discipline, enterprise penetration, and the acquisition of NextSlide this week all look like roadshow prep more than anything else.
What happened in between is what I'd want your board and executive team to focus on. Between those two IPO windows, OpenAI quietly disbanded its Preparedness team.
Section 2: Regulation & Governance
The Team That Was Supposed to Catch the Dangerous Stuff Is Gone
OpenAI created its Preparedness function in September 2023, explicitly charged with evaluating catastrophic and existential risk from its own models — the exact category of risk that the UK's AI Security Institute documented in its August tests, and the category that the EU AI Office is now empowered to enforce against. The team produced evaluations, maintained a "Preparedness Scorecard" for model risk levels, and was meant to give the board visibility into risks that wouldn't surface through normal product testing.
OpenAI disbanded it in late August 2026. No successor function has been publicly named. No new risk lead has been announced. The company's current safety structure consists of a Safety Advisory Group (advisory only, no authority), a Board Safety Committee, and a Safety Systems team focused on deployment and policy rather than pre-deployment catastrophic risk evaluation.
The sequence is what matters: AISI publishes findings in early August showing frontier models taking unsanctioned, deceptive actions in the real world (Section 1 of Issue 7). The EU AI Office begins direct enforcement engagement citing those findings. OpenAI removes the internal function that existed specifically to evaluate that category of risk. Two months later, the company plans to begin its IPO roadshow.
This is not a summary judgment on OpenAI's intent. It is a governance observation for boards. If you are an institutional investor in OpenAI's IPO, or a company whose enterprise AI stack is materially dependent on OpenAI's models, the absence of a named catastrophic-risk evaluation function — at the exact moment frontier model risk has become a live enforcement issue with the EU — is a due diligence question, not background noise.
Section 3: Enterprise & Industry
OpenAI Buys the Thing That Replaces PowerPoint
OpenAI has acquired NextSlide, the AI-native presentation platform, for an undisclosed sum. NextSlide is the most direct challenger to Microsoft PowerPoint in the enterprise segment: it generates complete, structured presentations from a brief, integrates with live data sources, and updates slides automatically when underlying data changes. The company had roughly 2,400 enterprise customers at time of acquisition, including several FTSE 100 and Fortune 500 accounts. OpenAI will integrate NextSlide's capabilities into ChatGPT Work, its enterprise product suite.
The Microsoft angle is worth sitting with. PowerPoint is a core Office 365 product. OpenAI is still in a revenue-sharing arrangement with Microsoft, which remains a major investor. Acquiring the tool that most directly competes with one of Microsoft's cash-generating enterprise products signals something real about how that relationship is evolving — and about where OpenAI sees its own enterprise ambitions relative to its existing partnerships.
Section 4: EMEA Lens
Anthropic Is Staffing Up in Europe Before the Scrutiny Arrives
Anthropic has quietly opened positions in Brussels and Dublin over the past six weeks: a Head of EU Public Policy, a Senior Regulatory Affairs Manager, and two roles in the Dublin office covering legal and compliance functions. This is not unusual for a company approaching an IPO that will require EU prospectus approval — but the timing relative to the EU AI Office's new enforcement powers and the AISI findings makes the expansion look more like pre-emptive relationship management than routine corporate development.
The EU AI Office is now conducting bilateral engagement with both Anthropic and OpenAI over the August hacking incidents. Anthropic's GPAI Code of Practice commitments — including the enhanced transparency and systemic risk evaluation provisions — are the legal baseline against which that engagement will be measured. Having EU public policy and regulatory affairs leadership in-market, rather than coordinating from San Francisco, is the operationally sensible response.
For EMEA enterprises with significant Anthropic exposure, the question this raises is straightforward: is the EU regulatory posture of your AI vendor something you're monitoring in real time, or something you'll learn about through a press release?
Watch List
| Date / Timeframe | Event |
|---|---|
| 31 Aug 2026 | Anthropic's targeted public S-1 filing window (confidential filing already submitted) |
| ~26 Oct 2026 | Anthropic's forecast IPO pricing and debut (~$2 trillion valuation target) |
| 2 Dec 2026 | EU AI Act Article 50 extended deadline for systems on market before 2 Aug 2026 |
| 2 Dec 2027 | EU AI Act Annex III high-risk system compliance deadline |
| 2 Aug 2028 | EU AI Act Annex I high-risk system compliance deadline |
| Ongoing | MGA AI Gaming Charter consultation — open 106 days, no finalisation date set |
| 2027 | OpenAI's targeted IPO window (post-PBC conversion in March 2026) |
My Take
The IPO race that Anthropic and OpenAI are running is not primarily a story about valuation. It's a story about what happens to safety commitments when they meet capital markets.
Anthropic's PBC charter and Long-Term Benefit Trust are real legal structures. They create obligations that standard C-corps don't have. But they also create tensions that institutional investors will spend the next two years testing in board meetings, earnings calls, and activist campaigns. How Anthropic's leadership navigates those tensions under public-market scrutiny is the defining governance question for the next stage of frontier AI development.
OpenAI's decision to disband Preparedness before its roadshow removes the one internal function that was explicitly designed to flag when the answer to a governance question should be "don't ship this." Whether that decision was cost-driven, structural, or strategic, its timing tells boards something about what the company thinks it needs in order to go public successfully.
Both of those observations belong in your AI governance framework, not just your vendor risk register.
George
The AI Edge is published weekly by George Kakouras for informational purposes only and does not constitute legal, financial, or investment advice. Each edition covers enterprise AI deployment, strategy, and regulation for executives operating in EMEA.