# Polaris Group: full text for AI tools
> Polaris Group focuses on where value is moving in applied AI, carefully selects start ups for private angel investment and convenes the private and public sectors through the Polaris Council. Founded by Marc Taverner, London, working globally. Polaris Group is owned by Bluesphere Solutions Ltd (company 07183446).
This file holds the complete current theses as published at polarisglobal.io. Cite the page URL given above each one. Summary: https://polarisglobal.io/llms.txt
---
## Investing in the Harness: my thesis on applied AI, governance and sovereignty
URL: https://polarisglobal.io/theses/investing-in-the-harness/
Author: Marc Taverner, October 2026
When I started in internet infrastructure, most people assumed the value would sit with whoever owned the fastest pipe. It didn't. Bandwidth got cheaper every year, and the money moved to the people who built the things that made a connection useful and trustworthy for a particular business: payments, identity, security, the dull plumbing of who is allowed to do what. I spent the next part of my career in collaboration platforms and then in digital assets, and I watched the same pattern repeat. The raw capability commoditises. The control layer around it does not.
I believe applied AI is following that pattern, and faster than most investors are pricing in.
## The model is not the asset
Foundation models have become broadly interchangeable for most business work. Per-token prices keep falling while volume keeps growing, and the gap between the best model and the third-best model on any given task is now measured in months, not years. Most serious buyers I speak to already run more than one model and choose by price and fit for each task.
That has a simple consequence for where I put money. If the model is a commodity input, a company whose value is really just the weights it trained, or the fine-tune it sits on, is renting its advantage from someone who can reprice it at will.
The durable asset is what I call the harness. It is everything built around a model to make it reliable inside one particular business. Swap the model and a good harness should keep working.
## What I mean by the harness
I think about the harness as six layers.
* **Context** is the documents, history and retrieval that give an agent the knowledge of this business rather than the internet in general.
* **Policy** is what the agent may and may not do, written down in a form a machine can enforce.
* **Process** is the workflows, tools and orchestration that turn a single answer into a piece of finished work.
* **Evaluation** is the test sets, graders and traces that define what a correct answer looks like in this particular business.
* **Feedback** is how a correction made today improves the system tomorrow.
* **Identity** is who the agent is, who it acts for, and what it is entitled to touch.
These layers are not equally investable. Context, in the narrow sense of conversational memory, became a free platform feature on major platforms between September 2024 and March 2026. I would not fund it as a standalone product today, however good the team.
Evaluation and identity are a different story. The definition of a correct answer in an insurer's claims process, or a bank's onboarding flow, is specific to that organisation. It is also exactly the thing an auditor, a regulator or an insurer will ask to see when an agent gets something wrong. Identity is the question every security team is now being asked by its board: which agents are running, on whose authority, with which credentials, and who can switch them off.
What I want to fund is the governance, lineage and evaluation infrastructure that works across clouds and across model vendors, produces evidence that a regulator or auditor will actually accept, and is sold to the security and compliance buyer. One condition matters more than any other. The customer keeps ownership of its own evaluation content and test cases. That is the asset. The vendor's job is to make it usable, not to hold it hostage.
## Leasehold and freehold
There is a property analogy I keep coming back to. A fine-tune on a closed frontier model is a leasehold, not a freehold. OpenAI closed self-serve fine-tuning to new customers on 7 May 2026, and from 6 January 2027 existing customers can no longer create new fine-tuning jobs either. On Azure there is no way to export fine-tuned weights, and fine-tuned deployments are deleted after fifteen days of inactivity, or twenty four hours on the developer tier. The model itself is retained and can be redeployed, but you never hold the weights. You paid for the improvement, and you still don't own it.
A company's own evaluation suite, its written policies and its curated context are different. They sit on its own systems. It holds them by possession rather than by contract, which is the only form of real ownership this market currently offers.
Whether that harness genuinely survives a change of underlying model is something I am treating as a thesis to test, not a fact to assert. I think it does, mostly. I don't know yet how much rework is involved in practice. So I am measuring it myself. I am building my own personal AI agent, a genuinely personal assistant that does my digital work while keeping my memory, context and data under my control rather than captured inside one vendor's platform. Every so often I swap the underlying model and record two things: how much rework the change needs, and whether a fixed evaluation suite catches the resulting drift before it causes a real mistake.
## Who actually pays
Here is the commercial reality I have learned to respect. Individuals will not pay for privacy they can get for free, nor do the majority care about privacy beyond the utility it can deliver. For example, using a VPN to access content otherwise geolocation restricted. Roughly four in five password manager users sit on the free tier. VPN users are more willing to pay, and free use there is a fast growing minority rather than a majority, but the direction is the same. I like consumer privacy as a cause. I don't back it as a business.
Enterprises are different because they have obligations, not preferences. They have audit requirements. They have data residency law. They carry contractual liability for what an autonomous agent did on their behalf. Nobody buys a compliance control for fun; they buy it because the alternative is a fine or an uninsurable risk. That is the buyer I want the companies I back selling to.
Revealed enterprise spend on securing AI is still a small fraction of total AI spend. Some people read that as a ceiling. I read it as an early category. Cloud security, identity and endpoint protection all started as rounding errors on the budget of the thing they protected, and each became a large market once something went wrong publicly and the obligation caught up with the technology.
I also want to be honest about where the independent opportunity is narrower than the headline suggests. An enterprise that runs everything through one model vendor will simply buy controls from that vendor. The opening for an independent company is in three places. The first is enterprises running agents across several clouds and several model vendors, where no single vendor's tooling can see the whole picture. The second is enterprises with real jurisdictional exposure, where the incumbent vendor is itself part of the risk being managed. The third is any situation that needs evidence for a genuinely independent third party, such as an auditor or an insurer, who will not take a vendor's word about its own product.
That last point is the heart of it. Neutrality is the product.
## What actually works for privacy today
A great deal of money is being raised on privacy technology that does not yet do what the pitch says, at least not for AI inference.
Confidential computing works. It runs a workload inside a secure hardware enclave, so that even the cloud operator cannot see the data while it is being processed, and it can prove to the customer which code ran. It is in production at the major clouds and on current GPUs, and the performance overhead has become small. Keeping sensitive context on the client side and sending only what a task needs also works, and it costs nothing.
Fully homomorphic encryption and multi-party computation are not practical for AI inference at any meaningful scale today. They are serious research fields. They are not where I would put a privacy guarantee for a production agent in 2026.
Zero-knowledge proofs are the most frequently misapplied of all. A zero-knowledge proof shows that a computation was carried out correctly. It does not let a model compute on data it cannot see. When someone pitches me zero-knowledge privacy for AI, my question is simple: who holds the unencrypted data or supplies the trust protocols while the proof is being generated? The answer usually ends the conversation.
Simple tokenisation or data masking is not enough on its own either. Published re-identification studies show it can often be reversed.
What works today is a combination. Confidential computing for workloads that need cloud scale. Client-held context, so the sensitive material never leaves the customer's control by default. Local inference on open-weight models for the genuinely sensitive fraction of the work. Everything else runs on whichever frontier model is best and cheapest that month, and a cross-vendor evaluation suite governs the whole arrangement so the customer knows when quality moves. None of that needs a breakthrough. It needs good engineering and someone neutral to stitch it together.
## Europe is the long-dated opportunity
Most of what I have described is a race against the platforms' release schedules, and I am realistic about who usually wins that race. There is one opportunity that does not depend on winning it, because it is created by law and politics rather than by a product roadmap. That is European sovereignty.
AWS's European Sovereign Cloud reached general availability in January 2026. It runs as a separate cloud partition, operated through a German holding company and three German subsidiaries, with EU national managing directors and an advisory board of five EU residents. Three of those five are Amazon employees, and the customer contracting entity is still Amazon Web Services EMEA in Luxembourg, so this is corporate separation inside the Amazon group rather than independence from AWS. The first region is in Brandenburg, backed by a stated commitment of 7.8 billion euros in Germany through 2040, with named customers including EWE, the Medizinische Universität Lausitz, Sanoma Learning and, since May, SCHUFA.
That gap between the marketing and the ownership is not a reason to dismiss it. It is the argument for the neutral layer I want to fund, because a European enterprise that needs to prove where its data went cannot rely on an attestation from a subsidiary of the company it is trying to be independent of.
At the same time, Mistral AI is positioning itself as Europe's own independent sovereign AI champion while being a launch partner on that same American hyperscaler's sovereign cloud, and while holding a multibillion dollar partnership with Microsoft announced in July 2026 under which Microsoft draws on Mistral's European GPU capacity for its own cloud. Its largest outside shareholder is ASML. I don't see a contradiction there. I see a market still forming its coalitions, which is exactly when the infrastructure that sits between the parties gets built and gets valuable.
The legal pressure is durable. The EU Data Act's switching provisions have applied since September 2025. What lands on 12 January 2027 is narrower and sharper: switching charges go to zero altogether, and because the regulation defines switching charges to include data egress, egress for a mandated move goes with them. That makes it materially easier for a European enterprise to move workloads and materially harder for a vendor to lock them in. Egress for running two providers in parallel survives, at cost.
The EU AI Act creates real obligations around fine-tuning and deployment that a vendor based outside Europe cannot simply contract its way around. Fine-tuning can make you the provider of the model once the compute you spend on it passes roughly a third of the original model's training compute. Deployment can do the same if you rebrand a system, substantially modify it, or change what it is for. Those roles are allocated by law, not by contract.
These obligations do not get commoditised by a feature launch. They do move, though, and in 2026 they moved the wrong way for anyone underwriting enforcement. In July 2026 the Union deferred the high risk obligations by sixteen months under its simplification agenda, to December 2027 and August 2028. It left the general purpose AI obligations untouched. I am underwriting the general purpose limb, which has applied since August 2025, not the high risk timetable.
I don't want to fund another sovereign cloud. What I want to fund is the governance, lineage and evaluation layer that lets a European enterprise prove to a regulator that its agents ran where they were supposed to, under the law that actually applies, across infrastructure that mixes American hyperscalers, European model providers and open-weight models running on its own hardware. Proof is the product.
## Two shapes of company
I seek to back and work with two different shapes of company.
The first shape is near-term. These are companies built to be acquired by identity and security incumbents who need the capability quickly and cannot build it in time. Okta, CrowdStrike, Cisco, SailPoint and Cyera are the kind of buyer I have in mind.
Recent acquisitions of focused security and identity companies have mostly landed between one hundred million and one billion dollars. The median priced deal in these categories sits nearer three hundred million than a billion, and roughly nine in ten acquisitions close on undisclosed terms, which usually means smaller rather than larger.
The second shape is longer-duration. These are companies whose advantage is genuine neutrality across clouds and model vendors, or genuine jurisdictional specificity, most often European. They do not need to win a sprint against a hyperscaler, because the thing protecting them is law and neutrality rather than a feature lead. They take longer and need more patient capital. They are also the ones that can become durable independent businesses.
## The companies I find interesting
The first area is agent governance, identity and audit, and it has moved fast. Onyx Security is building controls for how AI agents operate inside the enterprise, so a security team can see and govern what those agents actually do. Keycard is working on identity and access built specifically for agents, the question of who an agent is and what it may touch, designed in from the start rather than retrofitted from human login systems. AIUC approaches the same problem from the insurer's side, setting a standard for AI agent risk and tying it to underwriting, which is precisely the independent third party I described earlier. GitGuardian built its name on detecting leaked secrets in code and has become a natural home for the credentials sprawl that agents create.
Two things that were on my list here have already gone, and both are the pattern rather than exceptions to it. Oasis Security came out of the non-human identity problem, the service accounts, keys and tokens that now outnumber human users in most organisations, and extended that into managing the access AI agents are granted. Cyera bought it in September 2026 for about a billion dollars. Cisco bought Galileo, an evaluation and observability company I would otherwise have named below, in May 2026 and folded it into Splunk.
These companies sit on an obligation rather than a preference, each is most useful when it sees across vendors, and each produces evidence someone other than the vendor will rely on. They have been funded at seed through Series C. None of them yet discloses meaningful revenue publicly. I say that as an observation, not a criticism. It tells me the category is still being underwritten on thesis and on team, which is when the best entry points exist and when judgement matters most.
Cyera and Cisco are consolidating this exact space quickly, Cyera having bought Otterize and then Oasis, Cisco having bought Astrix, WideField and Galileo. That is good for exit prices and bad for anyone who wants to build the second, longer-duration shape here..
Evaluation and observability infrastructure is the second area. Braintrust is a good example of what I mean: tooling that lets a company define what a correct answer looks like in its own business, trace every agent decision, and keep that evaluation suite as its own asset regardless of which model sits underneath. This is the layer that makes the leasehold and freehold distinction real for a customer.
Digital identity wallet middleware is the third. Every member state is required to offer an EU Digital Identity Wallet by 24 December 2026. As I write, one has. Denmark went into production in June. Germany, France and Italy are in public sandbox, no wallet is yet fully certified under the EU scheme, and Sweden is not expecting one before 2028. Someone has to connect those wallets to the agents acting on a person's or a company's behalf, and the gap between a binding deadline and a single live implementation is the opportunity rather than an argument against it. Procivis, the Swiss firm building wallet and credential infrastructure already used in government rollouts, shows the shape of the thing I mean, though it has been majority owned by Orell Füssli since 2021, so it is an illustration rather than a target. What I am looking for is the plumbing that lets an agent present verified credentials and act within a mandate the human has actually granted.
Post-quantum crypto-agility is the fourth. Every signed credential, every agent identity and every audit log I have described rests on cryptography that will need replacing over the next decade at least & maybe sooner. PQShield, the Oxford firm selling post-quantum tooling to enterprises and silicon vendors, is an interesting example: not a new algorithm, but the ability for an enterprise to find, inventory and swap its cryptography without rebuilding its systems. Regulators have already set migration timelines, so this is an obligation with a date attached.
Confidential computing and attested inference is the fifth. Anjuna Security is building the kind of product I described in the privacy section: running workloads, AI inference included, inside hardware-attested enclaves across any cloud, and giving the customer a cryptographic receipt showing which model ran, on which hardware, against which data. That receipt is exactly the sort of evidence an auditor can accept.
European sovereign AI infrastructure is the sixth. OVHcloud is the right shape not because it competes with the hyperscalers on scale, but because it gives European enterprises a credible, jurisdictionally clean option for the part of their work that has to stay under European law. Its agreement to acquire Dragon LLM, signed in March 2026, and the sales organisation it has built around sovereign demand shows that it is building the plumbing, not just the marketing.
Across all six, the common thread is the same. The company makes the harness portable, provable or protected, and it sells to someone who is obliged to buy.
## What I’m not too interested in
The following examples I find less appealing to engage with;
* Model companies, and anything whose value is really just the underlying weights.
* Anything relying on fully homomorphic encryption, multi-party computation or zero-knowledge proofs as the actual privacy mechanism for AI inference. I’ve no evidence they work at this scale yet.
* Anything whose entire differentiation is a feature already on a hyperscaler's public roadmap.
* Any company that wants to own the customer's evaluation data and test cases rather than letting the customer keep them. That is lock-in dressed up as a sovereignty pitch, and I won't back it.
## What would change my mind
There are two open questions I can't yet answer, and they are the two that would move me most.
The first is portability in practice. I have not yet found a single named enterprise that has publicly credited the portability of its own evaluation harness with letting it switch its primary AI vendor. I believe it is happening quietly. I would like to see it on the record. If, over the next year or so, large buyers start citing their own evaluation suites as the reason a model switch took weeks rather than quarters, the freehold argument becomes a proven commercial one. If switches keep happening and nobody credits the harness, I will need to rethink how much of this value really sits with the customer.
The second is who captures the spend. I don't yet have good data on whether enterprises buy this kind of governance tooling from independent companies, or simply absorb it into contracts they already hold with their incumbent security vendors. If independent companies keep winning standalone budget lines, the second, longer-duration shape gets much more attractive. Does the spend folds into existing Okta, CrowdStrike or Microsoft agreements?
A third has been added by events rather than by argument. The European regulatory timetable moved in 2026, and it moved toward simplification rather than enforcement. If the high risk deferral turns out to be the first of several, and the general purpose obligations are softened in the same way, the European limb of this thesis weakens considerably. I am watching the digital omnibus files, which were still unresolved in Council as I write.
## Where I have landed
My underlying convictions haven't changed since I first started writing about this. I care about portability. I care about sovereignty, at the personal level and at the national level. I care about privacy, and I want agentic AI that is genuinely useful and genuinely trustworthy, not one or the other.
What has changed is how precisely I can say where those convictions meet a cheque I'm willing to write. I back companies that own the harness rather than the weights, that sell to an obligation rather than a preference, and that compete on neutrality rather than secrecy. I back them in Europe where the law gives them a long runway, and anywhere an independent party needs evidence a vendor can't provide about itself.
If you're building one of those companies, or investing in them, or writing the rules they will have to prove they follow, I'd like to talk.
## Evidence notes
*This edition was checked against primary sources on 22 and 23 October 2026. Data and content is accurate to then as best I have been able to check. I am human, I may make mistakes & I may have thoughts and opinions you don’t agree with. These are likely driven by my own personal experiences and humanity.*
* **OpenAI fine-tuning.** OpenAI deprecations page, "Update to OpenAI's self-serve fine-tuning", announced 7 May 2026. Three stages: 7 May 2026, 2 July 2026, 6 January 2027. developers.openai.com/api/docs/deprecations
* **Azure fine-tuned deployments.** Microsoft Learn, Azure OpenAI fine-tuning deployment and cost management pages. Fifteen day inactivity deletion; twenty four hours on the Developer tier. No Microsoft document states that weights cannot be exported; there is simply no mechanism.
* **Password manager and VPN free tiers.** Security.org password manager annual report, 79 per cent on free apps, fieldwork October 2024. Security.org VPN consumer report, 43 per cent on free services, fieldwork August 2026.
* **AWS European Sovereign Cloud.** AWS News Blog 14 January 2026 and AWS press release, Potsdam, 15 January 2026. Governance and entity structure from aws.eu/governance and the AWS overview whitepaper. Investment figure of 7.8bn euros through 2040 from the AWS announcement of 15 May 2024, repeated January 2026.
* **Mistral.** AWS launch partner list, 15 January 2026. Microsoft and Mistral expanded partnership, 21 July 2026. ASML Series C, 9 September 2025, approximately 11 per cent fully diluted.
* **EU Data Act.** Regulation (EU) 2023/2854. In force 11 January 2024, applicable from 12 September 2025. Article 29(1) ends switching charges from 12 January 2027; Article 2(36) defines switching charges to include data egress. Recital 99 preserves egress charges for in parallel use, at cost.
* **EU AI Act.** Regulation (EU) 2024/1689, Articles 2, 25 and 113. Commission guidelines on the scope of GPAI obligations, C(2025) 5045 final, 18 July 2025, for the one third compute threshold. Regulation (EU) 2026/1744, in force 27 July 2026, for the deferral of Chapter III to 2 December 2027 and 2 August 2028.
* **EU Digital Identity Wallet.** Article 5a of Regulation (EU) No 910/2014 as amended by Regulation (EU) 2024/1183. Deadline of 24 December 2026, being 24 months from the entry into force of the first implementing regulations on 24 December 2024. Rollout status as at October 2026.
* **Acquisitions.** Cyera and Oasis Security, announced 28 July 2026, completed 3 September 2026, approximately $1bn, reported. Cisco and Galileo, announced 9 April 2026, closed 22 May 2026, terms undisclosed. Google and Wiz, $32bn, closed 11 March 2026. Palo Alto Networks and CyberArk, $25bn, closed February 2026. Median and disclosure-rate figures from SecurityWeek annual M&A reviews for 2024 and 2025 and from H1 2026 market reports.
* **GitGuardian.** Series C, $50m, February 2026, led by Insight Partners.
* **Procivis.** Majority owned by Orell Füssli since September 2021, following an initial stake in spring 2020.
* **OVHcloud and Dragon LLM.** Binding agreement announced 25 March 2026. No completion announcement had been published as at October 2026.
* **Platform memory.** Free tier availability: ChatGPT September 2024, Meta AI January 2025, Gemini March 2025, Copilot April 2025, Grok around April 2025, Claude around March 2026. The claim holds for conversational memory; richer retrieval features remain subscriber-gated on several platforms.
---
## Agents of Sovereignty: earning and owning value in an economy reshaped by AI, 2026 to 2031
URL: https://polarisglobal.io/theses/agents-of-sovereignty/
Author: Marc Taverner, The Polaris Group, updated 23 October 2026
## Where I'm starting
**I expect AI to change the economics of knowledge work quite dramatically. As more people can get the same work done, the value will increasingly sit with whoever brings the knowledge, expertise, relationship with the counterparty or customer and has the right to use the information. I want to back businesses that help individuals and companies put those advantages to work with AI, keep control of them and keep a worthwhile share of the value created.**
That's where Agents of Sovereignty starts for me. People need to find ways to remain useful and earn a living. Companies want the benefits of AI without handing over the knowledge and relationships their business depends on. Governments have a problem to work through too, because employment, tax receipts and the cost of supporting people are all tied up in this.
My bet is that as AI becomes easier to get hold of, knowing how to put it to work in a particular business becomes more valuable. There should be businesses worth building around that: helping a professional earn more, getting work into production that a company can't responsibly automate today, or making sure the customer can change tools without starting again. The return has to show up somewhere in the customer's economics. Otherwise we're selling a preference.
So the question I keep coming back to is, who actually does better when this product works? And why will they keep paying the company providing it?
## Work is changing, even before the jobs disappear
I think AI will change employment substantially, but redundancies are only part of it. Who gets hired in the first place, how a junior learns the job, what a client will pay and how much work a small team can take on. The job title may stay much the same while the economics underneath it change.
The evidence already warrants attention, although it does not justify a confident date for mass unemployment. Stanford's August 2026 analysis finds a widening employment gap for young workers in AI-exposed occupations, operating mainly through reduced hiring. The authors also find no widespread economy-wide displacement and describe the results as descriptive indicators rather than causal estimates. Danish research provides a useful counterweight: it finds changes in tasks and occupational movement without detectable changes in earnings or recorded hours over the period studied. [\[1\]](https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/)[\[2\]](https://bfi.uchicago.edu/wp-content/uploads/2025/04/BFI_WP_2025-56-4.pdf)
I don't think we should explain away that difference in the evidence. We don't yet know the pace, or which places and occupations will feel it most. But waiting for an unemployment crisis doesn't strike me as a sensible investment strategy either. A lot of the commercial relationships that determine who benefits will be settled before then.
The immediate opportunity lies in adaptation. Businesses will experiment with smaller teams, different roles and new ways of delivering services. Individuals will need to understand which parts of their contribution remain valuable and how AI can extend them. Some will prosper inside employment. Others will build businesses, combine several sources of income or find that their previous work commands a lower price.
I want to back businesses that work through that uneven transition. Some may benefit from rapid displacement, but I'm not comfortable making mass unemployment a condition of the whole thesis.
## Doing more work and getting paid for it
There's a fairly basic issue here, producing more doesn't mean you get to keep more.
AI can help someone do more work while leaving them no better off. Their employer keeps the saving, a client expects a lower price, or the platform takes its fee. And some of the apparent gain disappears into checking the output and finding the next customer.
Research on customer-support work illustrates both the opportunity and the distribution question. A field study found substantial productivity benefits, particularly for less experienced workers, with suggestive evidence that the system helped spread practices associated with stronger performers. That establishes a way to improve work. It leaves open how the resulting gains should be shared between the people whose experience informed the system, those using it, the employer and the supplier. [\[3\]](https://www.nber.org/papers/w31161)
For an individual, I want to know what's left after paying for the tools, checking the work, making corrections and winning the customer. For a company, I want to see the whole job. Five minutes saved in a demonstration is easy to like; five minutes saved at the cost of an hour's rework is another matter.
This is why I've moved away from universal income funded by ordinary personal data as the foundation of the thesis. I haven't seen enough evidence that people care deeply enough about the sovereignty of their data or their data is capable of producing meaningful, recurring income. I still care about people sharing in the value created from their lives and work. I'm less convinced by the payment mechanism I originally attached to that ambition.
My stronger starting point is expertise applied to a real customer need. A person may hold specialist judgement, a trusted relationship, access to a particular market or information that is difficult to reproduce legitimately. AI may let that person serve more customers or offer a service that was previously uneconomic. The opportunity depends on preserving something scarce as the cost of producing generic work falls. People without those advantages also need a route into earning; that is where access, learning and credible evidence of capability become important. Their need is real even where the payer has yet to be established.
I'm still interested in licensing expertise and data where someone has the right to supply it and a buyer has a reason to pay again. The same goes for services delivered by people using agents. There needs to be a customer at the other end.
There are counterexamples worth taking seriously. US Trump Accounts give eligible children a route into invested capital, including a $1,000 government contribution for qualifying US citizens born in 2025 to 2028, which a parent has to claim rather than receive automatically. On 19 October, Musk pledged one SpaceX share per American child next year. Details as yet remain unsettled. \[12\]\[13\]
Anthropic's announcement is a separate proposal: 5% of its shares would go into a public-benefit fund, subject to stockholder approval. It could support donations or distributions to particular populations. That isn't a promise of shares for every young person, and its eventual reach is still to be set. \[14\]
I welcome the direction. Owning a stake can give people a share of future growth, and that is a real counterexample to the idea that all the value must stay with the founders and platforms. I wouldn't dismiss the benefit to a family.
But I see these as partial responses to the immediate pressure. The US children's schemes are geographically bounded and don't answer the position of adults elsewhere whose work is changing now. Shares may compound for years, so the asset itself needn't be short term. What's missing is a settled, continuing entitlement tied to the value people contribute.
My concern is the bargain we might gradually come to accept. A small allocation today could prove a very low price for society's acceptance of decades of commercial use of our digital selves: our data, habits and accumulated judgement. Cheap, if it comes to stand in for a continuing claim on that value. I'm not saying these schemes purchase perpetual data rights; the announcements cited here do not establish that exchange. I'm saying we shouldn't let an attractive gift settle the much bigger question of who controls that use and who keeps getting paid.
## What I mean by sovereignty
Sovereignty is a big word. For me it comes down to fairly practical questions: what can I authorise, what can I refuse, what can I inspect, and what can I take with me if I leave? Then, what are those rights worth in practice?
For an individual, that could mean keeping a useful record of their own expertise, directing an agent acting on their behalf, deciding what information it may disclose, and receiving agreed compensation for a legitimate use of their contribution. It does not imply ownership of everything they encounter at work.
For a company, it could mean controlling access to confidential information, defining the authority of agents and staff, retaining the knowledge needed to operate a critical workflow, and having a workable alternative if a supplier changes its service or terms.
These interests will sometimes conflict. An employee's general experience, an employer's confidential process and a client's restricted information can sit inside the same conversation. A useful product must respect those distinctions. Putting the conversation in a private store does not settle the rights to reuse it.
Consider a professional who teaches an agent how to handle difficult customer situations. If the company retains all the resulting capability after that person leaves, the system may have strengthened the company's position while weakening the individual's. That can be a legitimate contractual outcome, but it cannot be presented as evidence that everyone has gained sovereignty.
Where a business claims to improve individual participation, I will look for the mechanism: higher retained earnings, enforceable control over a contribution, portable capability, ownership or an agreed share of proceeds. A consent screen alone will not satisfy that test.
I need to keep the commercial return and the individual's position separate. An enterprise product can be a good investment without making life better for the people working there. Fair enough, but let's say so. If this thesis ends up only protecting companies, it will have lost the concern that started it.
Some arrangements may work better through professional bodies, employers, unions or other collective institutions. Their role is worth investigating because a scattered group of individuals may struggle to negotiate terms that an organised counterparty can secure. The commercial case still needs a funded buyer and a useful service.
## Where companies will spend
It's easy to see how this starts inside a company. Someone finds an AI tool that helps them get their work done and starts using it. From their point of view it makes sense. The company then has to work out what information went into it, what the tool was allowed to do and who is responsible if it goes wrong.
The concern is supported by evidence. In the UK Business Data Survey 2026, 73% of businesses handling digitised data said they were uncomfortable with that data being used to train external AI models. This is a finding about attitudes to training; it should not be stretched into a refusal to use external services. Separate UK research identifies data security and accuracy as concerns while finding that concerns often do not prevent adoption. [\[4\]](https://www.gov.uk/government/statistics/uk-business-data-survey-2026/uk-business-data-survey-2026)[\[5\]](https://www.gov.uk/government/publications/ai-adoption-research/ai-adoption-research)
That's the opening I want to investigate. Work where AI is useful enough to be worth adopting, but the customer needs better controls before they can put it into everyday use. The value is in getting the work done reliably.
The first buyer may be the person responsible for the business outcome. Security, legal and compliance may determine whether the purchase can proceed. In other cases, a specific control failure will make the security buyer the budget holder. I would establish that relationship in each market rather than assume that every opportunity belongs in a CISO's budget.
The threat model also needs to be precise. Use of information for model training, disclosure to another user, excessive access by an agent, an unauthorised action and dependency on one supplier are different problems. A contractual restriction can address one without resolving the others. Controls need to match the actual risk and the consequences of failure.
An independent supplier has to win this argument against products the customer already buys. Microsoft is extending Agent 365 across external platforms, with registry connections to AWS and Google in public preview. Independent evaluation products already offer capabilities such as testing, tracing and self-hosted deployment. The market contains capable competitors with distribution. [\[8\]](https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/)[\[9\]](https://www.langchain.com/langsmith/evaluation)
A startup needs a reason to win: a job the incumbent handles badly, evidence the customer can't get elsewhere, or an integration that makes the whole thing practical. Being independent may help. It won't, on its own, get a purchase order signed.
## The harness, and what the customer actually keeps
I beleive the harness matters. I mean the knowledge, instructions, permissions, tools, examples and corrections that make a model useful for a particular job. Much of the work goes into getting those things right.
I want the customer to retain the parts that contain their knowledge and define acceptable work. I also want evidence that those parts remain useful as the underlying technology changes.
Having a copy of the files is a start. What I'd want to see is the customer moving the work to another system and still getting an acceptable result, at a cost they can live with. Models behave differently, integrations break and instructions go stale. We need to try the move.
Evaluations are valuable because they give the customer a way to make that judgement. They require maintenance, independent test cases and checks against actual business results. An evaluation suite can preserve an old mistake just as efficiently as a useful standard.
There may be a good recurring business in keeping all this working as the tools change. There may also be a lot of consulting dressed up as software. Before putting money behind it, I want to know how much of the work repeats from one customer to the next and what support costs do to the margin.
Customer ownership is a product requirement. The supplier's commercial advantage must come from continuing to deliver something worth buying.
## Government has a stake in this
Governments can't ignore this if employment becomes less secure, young people struggle to get started or more income moves away from labour. Tax receipts and support costs come into it pretty quickly. Equally, higher productivity could leave governments with more resources. It depends who gets the gains, how they're taxed and whether people find another way to earn.
Policy could influence that outcome through training, procurement, competition, employment protections, social insurance and taxation. It could also strengthen the position of existing technology suppliers. A policy response is not automatically a market for the businesses I would like to see built.
We can already see governments trying to help people adapt. The UK's AI Skills Boost programme targets 10 million workers by 2030. The IMF discusses training, social protection and tax changes. Useful signs of interest, although they don't amount to a commitment to personal data income or portable agents. \[6\]\[7\]
For investment purposes I'd put a funded programme a long way ahead of a speech about what ought to happen. A business relying on the government needs to know who can buy, where the money comes from and how long it will take to get paid.
The same discipline applies to regulation. We need to establish which obligations attach to the customer's actual activity and what would satisfy them. Neither a broad appeal to the AI Act nor an assertion of data ownership is enough. European guidance makes clear, for example, that modifying a general-purpose model does not automatically create all the obligations of a model provider. Data Act switching provisions also need reading against the specific service: the data a customer can take with it excludes anything protected by intellectual property rights or held as the provider's trade secret. [\[10\]](https://digital-strategy.ec.europa.eu/en/faqs/guidelines-obligations-general-purpose-ai-providers)[\[11\]](https://digital-strategy.ec.europa.eu/en/factpages/data-act-explained)
I see a useful role for policy work in improving the conditions for participation: clearer rights, credible evidence of skills, fairer terms for contributions and practical routes back into productive work. I would keep that work connected to commercial evidence, while judging its success separately from the return on an equity investment.
## Where I would put my time and money
I'd start with businesses helping customers do valuable, repeat work with controls they can rely on. A clear buyer, a useful result and a reason the existing tools aren't enough. I'm more interested in getting one job right and expanding from there than funding a broad platform while it searches for a customer.
Alongside that, I want to investigate businesses that increase the ability of individuals and small firms to earn. The attraction is a visible connection between useful capability and income. The test is demanding: repeat customers, retained earnings and a reason the service remains valuable when competitors have access to similar models. Tools that produce more generic output will have to overcome a difficult pricing problem.
I'm also interested in what survives a change of tools or working relationships: the tests, permissions, knowledge and evidence that let someone keep doing useful work. There has to be a real switching problem or cost behind the demand. A sensible standard still needs people to adopt it and someone to pay.
I would pursue workforce participation and compensation arrangements through smaller experiments and institutional relationships until a repeatable buyer emerges. This keeps the individual's position in the thesis without pretending that an unresolved market is already ready for venture capital.
Across these areas, I would look for access to customers, depth in a workflow, dependable execution and an advantage that improves with use. Any claimed data advantage must be based on rights the company actually holds. Any network effect must show how one participant makes the product more valuable to another; a growing database alone is insufficient.
The technology choice follows the problem. Privacy-preserving techniques, local processing, identity systems and shared ledgers may each be useful. Blockchain has a case where several parties need a shared record or settlement mechanism and cannot rely comfortably on one operator. It still has to beat a simpler arrangement on cost, usability and enforceability. A token does not create demand for the underlying contribution.
I will pass on businesses whose income case relies on ordinary personal data acquiring an unexplained price, that promise protections they cannot enforce, or whose principal feature is already adequate inside the customer's existing bundle. I will also pass on a service business priced as software without evidence that delivery can become repeatable. In every category, the business must operate within rights genuinely agreed with the people and organisations whose contributions it uses.
## Over the next five years
In the first year I'd spend most of the effort proving demand in a few specific workflows. Paid trials, a proper comparison with what customers already use, and all the implementation costs included. Building something for myself will teach me a lot. It won't tell me whether an enterprise will buy it.
In years two and three I'd put more time and capital behind the businesses where customers come back, delivery costs hold up and the product earns a bigger place in their work. We'll also have seen how the incumbents respond. If the agreed tests keep failing, I don't want to fund another explanation for why demand is just around the corner.
In years four and five, the question becomes whether the company can maintain its position through changing models, platform terms and regulation. Expansion, acquisition or liquidity may be possible. None should be assumed simply because the thesis has a five-year horizon.
At every stage I would test three futures. Under slower adoption, the company must justify itself through present-day productivity or control. Under uneven disruption, it must serve the occupations and businesses experiencing the change. Under rapid displacement, it must show that its buyer retains the ability to pay. A consumer business can face its strongest social need at precisely the moment its customers lose purchasing power.
Time and capital need different decisions. Some research, policy work or advisory activity may be well worth doing without becoming a venture investment. I don't want the thesis to force everything I care about into one return model.
## What would make me change my mind
I have to be willing to be wrong about this. Otherwise it's a story I like, rather than something I can use to decide where to invest.
If people use a product consistently and still earn no more after costs, that weakens the individual-income case. The same is true if customers and platforms take almost all the gain. Lots of usage isn't enough.
On the enterprise side, I'd worry if paid trials keep turning into custom projects, existing suppliers already do the job, or the checking costs eat the savings. A compliance concern needs to turn into a purchase at some point.
For portability, I require a real move between systems with acceptable outcomes and a measured cost. If customers rarely need to move, or can already move cheaply, the proposed premium will be difficult to sustain.
With government-dependent opportunities, I want to see support become funding, authority or an actual purchase. I'm not going to keep moving the date because I like the idea.
I will review these signals quarterly and revisit the full thesis annually. The important measures are retained income, customer outcomes, renewal, implementation effort, margins and the position of competitors. Funding announcements and model benchmarks provide context; they cannot substitute for those measures.
## Where that leaves me
I still believe the way we organise AI's use will have a lot to do with who benefits. The models matter, of course. So does who can use the knowledge, reach the customer and decide how the money is shared.
That's where I want to spend my time. Helping people find a stronger position from which to earn, and helping companies use AI without losing control of the things their business depends on. Sometimes those interests will line up. Where they don't, we need to be honest about it.
There should be good businesses here. I want to find the ones where the customer's gain is clear and the supplier can keep earning its place. That's the version of Agents of Sovereignty I'm taking into conversations with founders, customers and investors. Still plenty to test.
## Evidence notes
This version draws on the research discussed in October 2026, with the US equity and public-benefit announcements added on 21 October. Every cited source was checked against my primary document on 23 October 2026. I am human, I may make mistakes & I may have thoughts and opinions you don’t agree with. These are likely driven by my own personal experiences and humanity.
[\[1\]](https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/) Stanford Digital Economy Lab, Canaries in the Coal Mine?, revised August 2026. US payroll evidence; descriptive indicators, with limitations acknowledged by the authors.
[\[2\]](https://bfi.uchicago.edu/wp-content/uploads/2025/04/BFI_WP_2025-56-4.pdf) Anders Humlum and Emilie Vestergaard, Large Language Models, Small Labor Market Effects, September 2025. Danish adoption surveys linked to administrative records; limited to the population and period studied.
[\[3\]](https://www.nber.org/papers/w31161) Erik Brynjolfsson, Danielle Li and Lindsey Raymond, Generative AI at Work, NBER working paper, subsequently published in 2025. Evidence from a specific customer-support setting; it does not establish economy-wide income effects.
[\[4\]](https://www.gov.uk/government/statistics/uk-business-data-survey-2026/uk-business-data-survey-2026) UK Business Data Survey 2026. Attitudes to external AI training among businesses handling digitised data; not a willingness-to-pay study.
[\[5\]](https://www.gov.uk/government/publications/ai-adoption-research/ai-adoption-research) UK AI Adoption Research, updated February 2026, based on 2025 fieldwork. Reported adoption, benefits and barriers; the survey does not capture shadow adoption.
[\[6\]](https://www.imf.org/en/Publications/Staff-Discussion-Notes/Issues/2024/06/11/Broadening-the-Gains-from-Generative-AI-The-Role-of-Fiscal-Policies-549639) IMF, Broadening the Gains from Generative AI: The Role of Fiscal Policies, June 2024. Policy analysis and recommendations, rather than an enacted programme or forecast of certain unemployment.
[\[7\]](https://www.gov.uk/government/publications/ai-skills-boost-explainer) UK AI Skills Boost explainer, January 2026. Government-industry programme and target; not evidence that the target has been achieved or that participation increases earnings.
[\[8\]](https://www.microsoft.com/en-us/security/blog/2026/05/01/microsoft-agent-365-now-generally-available-expands-capabilities-and-integrations/) Microsoft, Agent 365 capabilities and integrations, May 2026. Agent 365 itself is generally available; the AWS Bedrock and Google Cloud registry sync is in public preview, with lifecycle governance over those agents still to come.
[\[9\]](https://www.langchain.com/langsmith/evaluation) LangSmith evaluation product documentation. Evidence of an existing competing offering, rather than independent proof of performance or customer outcomes.
[\[10\]](https://digital-strategy.ec.europa.eu/en/faqs/guidelines-obligations-general-purpose-ai-providers) European Commission, Guidelines on obligations for General-Purpose AI providers. Modification makes an actor a provider only where it uses more than one third of the original model's training compute, and even then the obligations are limited to documenting the modification.
[\[11\]](https://digital-strategy.ec.europa.eu/en/factpages/data-act-explained) European Commission, Data Act explained. Switching and egress charges end on 12 January 2027; the exportable scope covers input and output data and metadata but excludes material protected by intellectual property rights or constituting the provider's trade secret.
[\[12\]](https://www.irs.gov/trumpaccounts) IRS, Trump Accounts. Eligibility and the $1,000 pilot contribution; not a universal payment to all children, and the contribution must be elected on Form 4547 even where Treasury has opened the account automatically.
\[13\] Associated Press, 19 October 2026. Musk's proposed SpaceX share gift; announcement rather than completed distribution. The pledge, its date and its unsettled details are corroborated by several outlets carrying the AFP report of 19 and 20 October 2026.
\[14\] Anthropic, Investing in the public benefit, 18 October 2026. Proposed equity commitment, approval condition and potential uses; distribution arrangements remain to be developed. The announcement also provides for a further 2% at the board's discretion for every 50% increase in company value above a baseline.