Simpson Associates Secures Beech Tree Investment to Accelerate Data & AI Growth

Simpson Associates’ announcement that it has taken a strategic investment from Beech Tree Private Equity marks a clear inflection point for a UK‑based Microsoft‑aligned data transformation firm — one that gains scale and capital while raising familiar questions about governance, productisation and operational continuity.

Background​

Simpson Associates is a York‑headquartered data transformation consultancy with an additional office in Sheffield, positioned as an end‑to‑end partner for organisations in regulated sectors such as policing, healthcare, financial services, education and government. The firm offers advisory services, platform builds on Microsoft Azure and Databricks, and managed operations — a full life‑cycle consultancy model that emphasises governance, security and production readiness.
The company publicly announced on 24 October 2025 that it has accepted investment from Beech Tree Private Equity; Simpson frames the capital as growth funding to accelerate organic expansion, broaden capabilities — explicitly including agentic AI — develop sector‑specific products and pursue strategic acquisitions. The press release names Giles Horwood as CEO, Rachel Hillman as CFO and Darren Moors as CRO, and reports the business employs “over 100 data and AI professionals.”
Beech Tree’s stated remit — investing typically between £10 million and £40 million into fast‑growing, profitable businesses in technology, tech‑enabled services and financial services — makes this a logical partner for a mid‑market, profitable services business looking to scale through productisation and bolt‑on M&A.

Why this matters now​

The market for regulated‑industry data platforms is moving decisively from proofs‑of‑concept to production‑grade deployments. Buyers in policing, health and finance increasingly demand demonstrable data lineage, security controls, auditability and managed operations — attributes that favour partners who can deliver strategy, engineering and 24×7 run‑time support. Simpson’s Microsoft and Databricks alignments, combined with specialist public‑sector experience, position it well for that demand curve.
Private equity participation in this segment follows a clear industry logic: PE firms can inject growth capital, professionalise operations (finance, HR, commercial), and fund M&A to buy missing capabilities fast. For customers this can mean more productised offerings and deeper delivery benches; for employees and customers it also means incentives and priorities can shift quickly.

---

What the announcement actually says​

• Simpson Associates announced the Beech Tree investment on 24 October 2025 and stated the capital will be used to accelerate organic growth, broaden capabilities in agentic AI, expand sector offerings and pursue strategic acquisitions.
• Simpson reiterated its credentials as a Microsoft Solutions Partner with several Azure specialisations and as a Databricks and IBM Cognos partner; it also highlighted its 2024 Microsoft Partner of the Year award for Community Response.
• The press release does not disclose headline transaction value, ownership percentages or detailed governance terms post‑deal — an omission that’s common in mid‑market PE announcements but materially important to stakeholders.

The explicit priorities spelled out by Simpson​

• Accelerate organic growth (sales and delivery capacity).
• Broaden service capabilities in emerging technologies, specifically naming agentic AI.
• Expand sector‑specific product offerings and scale managed services.
• Pursue strategic acquisitions to complement in‑house capabilities.

---

Verification and independent confirmation​

The primary announcement was distributed via GlobeNewswire and reproduced by Simpson’s own communications. Simpson’s Microsoft Partner recognition (Community Response Partner of the Year 2024) and its Solutions Partner specialisations are documented on its corporate pages and in prior press releases. Beech Tree’s investment profile and typical cheque size are publicly described on Beech Tree materials and in its recent deal announcements. Taken together, these independent sources corroborate the core factual elements of the deal and Simpson’s stated capabilities.
Where the announcement is intentionally silent — notably transaction size and precise ownership/governance arrangements — that absence remains a material unknown and should be treated as such by procurement teams and customers.

Simpson’s strengths going into this deal​

Simpson arrives at this moment with several tangible assets that make it an attractive PE platform:

• Microsoft pedigree — Microsoft Solutions Partner designations and multiple Azure specialisations reduce procurement friction on Azure‑centric tenders and provide practical technical guardrails for enterprise architectures.
• Regulated‑sector experience — established casework in policing and public safety (for example, TOEX) gives Simpson domain credibility where security and governance are non‑negotiable.
• Productisation in flight — Simpson already markets productised IP such as RedactXpert, an AI‑powered redaction tool used in policing workflows, which is exactly the type of asset private equity firms can scale into recurring revenue. Published case materials claim measurable outcomes from such products.
• Bench size — the firm publicly cites a headcount of “over 100 data and AI professionals,” which matters when bidding for multi‑force public‑sector frameworks and large enterprise deals.

---

Risks, guardrails and what customers should insist on​

Private equity can unlock growth, but it also changes the incentive architecture of a services business. For customers — particularly those in regulated industries — three categories of risk deserve immediate attention.

1. Governance and continuity​

PE deals frequently change board composition, approve faster roll‑outs of product roadmaps, and may change commercial terms as margin goals evolve. Customers should insist on:

• Named resource commitments and a staffed bench with backfill guarantees.
• Contractual continuity and change‑of‑control provisions.
• Clear SLAs and escalation paths for mission‑critical services.

2. Security and compliance posture​

When a partner accelerates work in agentic AI and other autonomous tooling, the attack surface and regulatory scrutiny increase. Practical procurement‑grade requirements include:

• Up‑to‑date penetration tests and SOC/attestation evidence.
• Independent threat modelling and red‑team results for any agentic components.
• Immutable audit logs, data residency controls, RBAC and Azure AD integration for identity binding.

3. Talent and integration risk​

Rapid M&A and shifts to productised delivery can unsettle staff. Customers should demand:

• Retention plans for named delivery leads and key technical architects.
• Transition and knowledge‑transfer plans for any acquisition integration.
• Contractual remedies if service quality degrades during integration windows.

---

Agentic AI: opportunity and governance headache​

Simpson’s press release names agentic AI as a target area for investment. The term broadly denotes systems that can plan, execute multi‑step tasks and act with some level of autonomy, and it’s increasingly cited by vendors as a core differentiator. In regulated settings these systems can deliver major efficiency gains — for example, automating redaction workflows, evidence collation, or case triage — but they also create new risk vectors: unintended autonomous actions, explainability gaps, data exfiltration and liability questions.
For customers procuring agentic solutions, vendors must provide:

• Human‑in‑the‑loop controls and identity binding for any autonomous actions.
• Immutable, queryable audit trails and versioned decision logs.
• Clear escalation and rollback procedures.
• Independent safety and bias testing, plus documented mitigation strategies.

Until Simpson publishes concrete product roadmaps and third‑party attestations for agentic capabilities, references to agentic AI should be treated as strategic intent rather than certified technical capability. That caveat is important: claims about future agentic features are forward‑looking and need verifiable proofs before being accepted for production use.

---

Market context: why PE is circling mid‑market data consultancies​

Private equity is actively targeting mid‑market consultancies that combine:

• Strong vendor alignments (Microsoft, Databricks).
• Regulated‑sector domain expertise.
• Evidence of productisable IP and recurring revenue potential.

This combination creates a route to multiple expansion by shifting revenue towards managed services and SaaS‑style contracts and by consolidating small, complementary practice areas into a scaled platform. Beech Tree’s historical playbook and stated cheque size perfectly align with this model, giving Simpson both capital and strategic optionality.

---

What to watch next (the milestones that matter)​

• Public disclosure of transaction economics and governance terms — who owns what and how the board changes. This matters to customers assessing long‑term alignment.
• First wave of strategic acquisitions — examine acquired teams for cultural fit, technical alignment and client overlap. Poor fits will show up as delivery disruption.
• Agentic AI pilots with independent attestations — pilots should include third‑party security and fairness testing plus demonstrable, audited before/after metrics.
• Product commercialisation cadence for assets like RedactXpert — look for an explicit roadmap, SLAs and FinOps guardrails for cloud usage.
• Employee retention indicators — named‑resource commitments, published retention plans and stability in delivery leadership are signals of healthy integration.

---

Practical checklist for IT procurement and CISOs​

• Demand named delivery leads and a bench‑staffing commitment for the first 12 months post‑deal.
• Require current security evidence: recent pen tests, SOC attestations and any sector‑specific accreditations.
• Insist on contractual change‑of‑control protections and data portability clauses.
• For agentic AI pilots, stipulate stage gates: (a) closed sandbox, (b) human‑in‑the‑loop evaluation, (c) third‑party red team, (d) controlled roll‑out with monitoring.
• Include FinOps clauses: periodic usage reviews, cost caps and escalation mechanisms to avoid cloud bill surprises.

---

Strategic implications for competitors and the channel​

For regional Microsoft partners and boutique consultancies, Simpson’s new capital backing represents both a threat and a market signal. PE‑backed competitors can outspend on sales, accelerate product teams and assemble M&A‑backed capability stacks quickly. That changes the competitive landscape in three ways:

• Larger bids become feasible for PE‑backed players when staffing commitments, operational resilience and balance‑sheet strength are required.
• Productised IP becomes a differentiator and a valuation driver; consultants that remain pure services risk being disintermediated.
• Channel motions with Microsoft may accelerate if Simpson operationalises co‑sell and consumption‑driven partner programmes; Microsoft alignment materially shortens procurement pathways in many regulated tenders.

For customers, the net effect can be positive if PE capital is converted into durable product and repeatable delivery models — but that outcome depends on disciplined execution, careful M&A integration and a commitment to security and sector expertise.

---

Critical assessment — pragmatic optimism with clear guardrails​

The investment is strategically coherent: Simpson has the vendor credentials, sector casework and early product IP that make it an attractive roll‑up and scale target. Beech Tree brings the capital and operational playbook to convert bespoke consulting into recurring, productised revenue — an archetypal private equity growth story.
However, the deal also raises three cautionary flags that customers should treat as negotiation levers:

• Undisclosed deal economics and governance: ask for clarity on ownership and decision rights.
• Agentic AI is an operational and regulatory risk: require independent testing and explicit human oversight.
• Talent concentration and integration risk: demand named resource protections and retention plans.

If Simpson executes with discipline — preserving delivery quality, publishing robust roadmaps and securing independent security attestations for agentic features — the investment can accelerate useful, auditable Data & AI platforms for regulated organisations. If it does not, customers will face the classic PE‑driven tension between growth and operational continuity.

---

Bottom line​

Simpson Associates’ decision to accept Beech Tree Private Equity as a growth partner is a consequential, but predictable, step for a Microsoft‑aligned data consultancy with product ambitions and public‑sector traction. The deal gives Simpson the capital and optionality to scale its Data & AI offerings, pursue M&A, and accelerate investments in agentic AI — but it also transforms incentives and raises operational questions that customers and partners should immediately address through contractual guardrails and rigorous procurement diligence. The measure of success will be concrete: published governance terms, demonstrable security attestations, stable delivery teams, and verifiable customer outcomes as Simpson moves from a strong regional consultancy to a scaled, product‑centric Data & AI business.

---

---

Source: GlobeNewswire Leading Microsoft Data Transformation partner Simpson Associates secures investment to accelerate growth and enhance their Data & AI capabilities

 

[ChatGPT]

Microsoft and OpenAI have signed a new definitive agreement that keeps the companies tightly partnered through the next technological era — but crucially changes the exit conditions for an AGI milestone by insisting any AGI declaration be verified by an independent expert panel before OpenAI can unilaterally alter the partnership's terms.

Background / Overview​

The Microsoft–OpenAI relationship has been one of the defining commercial partnerships of this AI decade: Microsoft invested heavily in OpenAI, built deep product integrations (notably Copilot across Windows and Microsoft 365), and supplied Azure compute to train and run the models that power ChatGPT and related services. That baseline of cooperation has now been formalized into a new, definitive deal as OpenAI restructures parts of its organization into a public benefit corporation (PBC) and prepares for expanded fundraising and operational independence. The companies say Microsoft will hold roughly a 27 percent stake in the newly capitalized OpenAI Group PBC, a position Microsoft values at about $135 billion on an as‑converted diluted basis.
At the same time, the agreement tightens and clarifies several controversial elements that had been the subject of public debate — especially the so‑called “AGI clause” that previous reporting suggested could let OpenAI limit Microsoft’s access to certain IP or services if OpenAI determined it had reached AGI. Under the new deal, that declaration cannot trigger unilateral changes without external verification: an independent expert panel must corroborate any AGI claim before the contract’s AGI‑linked provisions can take effect.
This article explains what’s changed, verifies the deal’s core numbers and legal levers against multiple public announcements and reporting, and analyzes what this means for Microsoft, for OpenAI, for the Windows ecosystem, and for enterprises and regulators tracking the progress toward Artificial General Intelligence.

---

What the new agreement actually says — verified facts​

Below are the key, load‑bearing elements of the definitive agreement as described by Microsoft and reported by multiple independent outlets. Each point is backed by at least two independent public sources where possible.

• Microsoft’s stake: Microsoft will hold approximately 27% ownership on an as‑converted, diluted basis in OpenAI Group PBC, valued at about $135 billion post‑recapitalization. This figure was disclosed in Microsoft’s own announcement and echoed by major news outlets.
• Public benefit corporation and recapitalization: OpenAI’s for‑profit arm will operate as OpenAI Group PBC, with governance and equity allocated such that the nonprofit OpenAI Foundation retains meaningful oversight and stake. Microsoft publicly states it supports the board moving forward with formation of the PBC.
• AGI verification by independent expert panel: If OpenAI declares it has achieved AGI, that claim will now be subject to verification by an independent expert panel before AGI‑triggered contractual changes take effect. This is the most consequential change to the earlier, more subjective mechanism reported in prior agreements.
• IP and exclusivity windows: Microsoft’s IP rights for models and products have been extended through 2032, and Microsoft retains rights to certain research IP until 2030 or until the independent panel verifies AGI, whichever occurs first. Microsoft’s product/model IP rights expressly exclude OpenAI’s consumer hardware. These clauses are specified in Microsoft’s public blog post and summarized across trusted reporting.
• Compute and cloud terms: OpenAI has committed to purchase an incremental $250 billion of Azure services under the arrangement, but Microsoft no longer holds a right of first refusal to be OpenAI’s exclusive compute provider. The agreement also clarifies which API products remain exclusive to Azure and which products may be deployed on other cloud providers.
• Product and partnership flexibility: OpenAI can jointly develop some products with third parties; API products developed with third parties will remain exclusive to Azure, while non‑API products may be served on any cloud. Microsoft can independently pursue AGI development and use certain OpenAI IP under constrained compute thresholds if it attempts to build AGI from OpenAI IP pre‑declaration.
• Open‑weight releases and government access: The deal permits OpenAI to release open weight models meeting capability criteria and allows OpenAI to provide API access to U.S. national security customers regardless of cloud provider.

These are the principal terms that Microsoft and other outlets publicly confirmed in their initial announcements. Independent reporting from Reuters, The Verge, and the Financial Times corroborates the major numbers and the high‑level contours of the IP, AGI verification, and compute commitments.

---

What we can and cannot independently verify​

Journalism standard: for core numerical and legal claims, the safest route is to rely on the parties’ public statements and reputable reporting — then label any privately negotiated language or leaked interpretations as reported rather than proven. Below is a breakdown.
What is verifiable now

• The Microsoft blog post summarizing the new agreement and its bullet points is authoritative about Microsoft’s stated position and the terms they publicly acknowledge.
• Major outlets including Reuters, The Verge, and the Financial Times independently reported the 27% / $135 billion figure and the broad contours of IP and compute commitments.

Claims that deserve caution or remain unverifiable

• Any precise contractual text, internal definitions, or side letters that were not published remain confidential; reporting about nuanced legal triggers (for example, the exact legal standard the expert panel will use to verify AGI) should be treated as reported summaries unless the full contract is released. The exact composition, remit, and decision standard for the “independent expert panel” are not publicly available in full text at the time of initial reporting. Treat claims about how it will rule as speculative until the panel’s charter or members are disclosed.
• Assertions that OpenAI attempted to plan a fast or premature AGI declaration to sever ties with Microsoft are based on investigative reporting and anonymous sources; those reports are important and worth scrutiny, but they are not legally proven facts. Use caution and label them as allegations or reporting when discussing them.
• Market capitalization valuations reported in some outlets (e.g., headline values like “OpenAI now valued at $500 billion”) can be estimates tied to the recapitalization math and investor rounds; they are useful guideposts but not regulatory valuations unless audited and publicly filed. Treat company valuation estimates as model‑based approximations rather than a single settled fact.

---

Why the independent expert panel matters — practical and governance implications​

The independent expert panel is the most consequential procedural change in this agreement. Here’s why:

• It replaces a board‑centric or unilateral trigger with a third‑party verification step. That reduces the risk of a conflict‑escalating unilateral declaration or a purely contractual “escape hatch” triggered by promotional or opportunistic benchmarks.
• It introduces a potential standardization point for what will count as AGI in commercial contexts. If the panel publishes a charter, test protocol, or a set of benchmarks, that could become an industry reference for other contracts and regulators.
• It creates a governance lever for Microsoft and other stakeholders to contest or accept AGI claims on a shared evidentiary basis, rather than litigating in court or relying on subjective board pronouncements. That reduces immediate legal uncertainty but adds dependency on the impartiality and competence of the selected experts.

Practical caveats

• The value of an expert panel depends entirely on who sits on it, how they are appointed, what evidence they review, and whether their determinations are binding or merely advisory. Those details are not yet public. Until the panel’s charter and membership are published, the panel is a conceptual fix rather than a demonstrated safeguard.
• Independent panels can themselves become politicized or captured by dominant industry viewpoints unless appointment rules and transparency safeguards exist. The risk is real: an expert panel could become a focal point for lobbying or legal challenge if its members or process are not clearly protected. This is a governance design problem that the broader AI policy community will watch closely.

---

What this means for Microsoft and the Windows ecosystem​

Short answer: Microsoft retains an extended and meaningful role, but it also hedges against being singularly dependent on OpenAI.
Microsoft’s strategic wins

• Product continuity: Microsoft keeps its role as OpenAI’s “frontier model partner,” preserving integration advantages for products like Copilot for Windows, Microsoft 365, GitHub, and Azure AI services. That continuity matters to Windows users and enterprises that rely on Microsoft’s bonded stack.
• Extended IP and product rights through 2032: These rights give Microsoft a long runway to monetize integrations and maintain competitive product features in Windows and Office even as OpenAI gains structural independence.
• Optionality to pursue AGI independently: Microsoft can now invest in its own AGI pursuits with third parties or internally — an explicit “insurance” policy against future divergence. Given Microsoft’s scale, Azure assets, and engineering teams, that is material leverage.

Risks and pressure points for Microsoft

• The loss of Right‑of‑First‑Refusal for compute reduces the exclusivity Microsoft previously enjoyed. If OpenAI chooses other cloud partners for future large training runs, Azure could miss some infrastructure revenue opportunities despite the $250B commitment. The economics of that tradeoff will be closely analyzed by investors and Azure customers.
• The extended IP window (2032) still telegraphs a finite time horizon. If OpenAI’s business model and ecosystem shift dramatically before then (for instance, with open weights or cross‑cloud distribution), Microsoft will have to accelerate product differentiation to retain customers.

What Windows users should expect

• Incremental, visible AI improvements in Windows and Microsoft 365 will continue in the near term because product integrations remain protected by the agreement. Expect continued Copilot evolution and deeper local/cloud hybrid scenarios for Windows features.
• Longer‑term, Microsoft’s ability to keep the best OpenAI models exclusively on Azure is constrained by AGI and by OpenAI’s greater multi‑cloud freedom. That means Windows may increasingly be an ecosystem where Microsoft competes with multiple top models rather than relying solely on a single OpenAI feed.

---

What this means for OpenAI and the broader AI market​

OpenAI’s strategic wins

• Recapitalization and PBC status give OpenAI more options to raise capital, reward employees, and align mission oversight with philanthropic commitments — while preserving operational agility for rapidly iterating model development. That structural flexibility is important if OpenAI needs to scale compute, offices, and research teams.
• Multi‑cloud operational freedom reduces single‑vendor risk and helps OpenAI manage bottlenecks in compute capacity and supply chains for high‑performance GPUs. Stargate‑style infrastructure plans and third‑party partnerships were reported prior to this deal; multi‑cloud access formalizes that option.

Risks and new responsibilities for OpenAI

• The independent panel requirement reduces the possibility of a unilateral “escape hatch” but also raises the bar for cleanly declaring AGI. OpenAI must now build transparent, auditable evidence if it ever intends to trigger AGI clauses — and that will require rigorous documentation and public communication strategies.
• The combination of open weights, third‑party partnerships, and national security API access expands OpenAI’s responsibilities for model safety, provenance, and compliance. Regulators and customers will expect clear control and red‑team results as models become more capable or widely distributed.

Market effects beyond Microsoft and OpenAI

• The deal accelerates multi‑cloud dynamics in the AI infrastructure market; other cloud providers (AWS, Google Cloud, Oracle, specialized hosts) stand to benefit from non‑exclusive compute opportunities. That will likely intensify competition on price, data center specialization, and bespoke hardware partnerships.

---

Strengths, weaknesses, and strategic risks — a critical assessment​

Strengths of the new deal

• Balancing innovation and governance: The independent panel provides a practical governance mechanism to defuse a major source of controversy — the AGI trigger — and could become a template for future deals involving transformative capabilities.
• Commercial and product continuity: Microsoft’s extended IP rights and product exclusivity protect existing customers and provide time to adapt if the partnership evolves.
• Operational flexibility for OpenAI: Multi‑cloud freedom and recapitalization help OpenAI scale without being fully captive to Azure’s capacity constraints. That reduces the execution risk for rapid model iterations.

Weaknesses and risks

• Opacity around the panel: The protective value of the independent panel is only as good as its charter and membership. Without transparency, the panel could create new legal and political battles. This remains the single largest governance vulnerability in the arrangement.
• Regulatory and antitrust scrutiny: The deal keeps huge concentrations of model ownership and commercial ties in corporate hands. That will attract regulators scrutinizing competition, national security access, and export controls. The more OpenAI releases open weights and broadens partnerships, the more complicated the regulatory picture becomes.
• Execution risk on the $250B compute commitment: Building, allocating, and delivering on multi‑year compute purchase commitments involves geopolitical, supply chain, and energy constraints. If compute economics change dramatically, the practical enforceability and benefit of the commitment could be renegotiated or strained.

Strategic wildcards

• If other major vendors accelerate in‑house AGI work (or if alternative architectures make AGI on commodity hardware feasible sooner than expected), the balance of power could shift faster than the 2030/2032 windows suggest. That’s the core strategic uncertainty underpinning all these negotiations.

---

Practical guidance for IT leaders and Windows administrators​

• Review and update identity and data governance policies now: agentic models integrated into workflows increase the importance of least privilege and explicit consent for automated actions.
• Audit integrations: inventory where Copilot and OpenAI model outputs feed automation; apply monitoring and logging to capture model decisions and downstream actions.
• Prepare vendor diversity plans: given the multi‑cloud turn, design cloud‑agnostic architectures for critical AI workloads where regulatory or cost constraints might require switching providers.
• Invest in explainability and red‑teaming: mandate model evaluation and adversarial testing before deploying high‑impact automation in production.
• Stay alert to policy/contract changes: the presence of an independent panel and long IP windows implies future legal and compliance checkpoints; ensure legal and procurement teams track public disclosures from both Microsoft and OpenAI.

---

Final assessment — why this matters to Windows users and the industry​

This definitive agreement is pragmatic: it locks in the commercial relationship that has driven product innovation in Windows and Microsoft 365 while addressing a major governance concern by introducing an independent verification step for AGI. It preserves product continuity and revenue for Microsoft, gives OpenAI more operational freedom to scale, and introduces a governance mechanism that, if run transparently, could set an industry precedent for handling transformative AI milestones.
At the same time, the arrangement does not eliminate key unknowns. The independent panel must be chartered and staffed publicly to deliver on its promise; the $250 billion compute commitment, while headline‑grabbing, will be complex to execute; and regulatory bodies will continue to scrutinize concentrated AI capabilities and partnerships.
In short: the deal reduces the chance of a sudden commercial rupture at the moment a model claims AGI, but it replaces that rupture risk with new governance and operational tests. For Windows users and IT leaders, the immediate takeaway is continuity with a guarded expectation of increasing model diversity and the need for stronger governance and operational readiness as capabilities scale.

---

Note: reporting and commentary circulating in industry forums and news outlets raised additional claims about OpenAI’s prior fundraising ambitions, internal discussions, and valuation metrics. Those matters are reported by journalists and industry commentators and should be treated as reported rather than fully verified legal facts until contract texts, panel charters, or SEC‑style filings become public.
For additional context and early community reaction, readers referenced recent forum and investigative threads that catalog the history of the AGI clause, prior negotiations, and speculation about compute constraints; those archives provide useful color but do not substitute for primary documents.
Conclusion: the Microsoft–OpenAI relationship enters a more formal, mapped phase — still cooperative, slightly less exclusive, and governed by a new layer of external verification that will be the crucial test of whether corporate guarantees can keep pace with a technical trajectory that many call unpredictable.

---

Source: Windows Central OpenAI won't sever its ties with Microsoft, even after declaring AGI — unless an independent expert panel verifies the claim