Soft Power Only Counts When You Spend It
The refusal that can change how we think about influence
Soft Power spent on a Hard Problem doesn’t fix the problem. It shifts the terms on which the problem gets fought. Which companies, funds, alliances, and multilaterals are actually doing it - and is yours one of them?
In February 2026, Anthropic CEO Dario Amodei told the Pentagon something rarely said by an executive: we cannot in good conscience agree to your request.
The request was to relax AI safety guardrails - specifically the ones that prevent Claude from being used for mass domestic surveillance and fully autonomous weapons. The Pentagon wanted them gone. Anthropic said no. Trump’s administration responded by labelling Anthropic a supply chain risk and ordering federal agencies to phase them out.
A reported nine-figure contract was gone.
That decision has been on my mind. It’s somewhat unusual in 2026 for a company to have values. And, it’s very much unusual for a company to spend them. At a real cost. Under real pressure. In public.
So, I started building an index.
Soft Power is only real when you spend it
Soft Power: everyone claims to have it. Very few spend it on anything that matters.
Reputation, restraint, governance, and trust - these are some of the currencies of Soft Power. They accumulate slowly, through consistent behaviour over time. And like any currency, they’re only worth something when you’re willing to exchange them for something real.
Most institutions and leaders spend their Soft Power on themselves. On panels. On positioning. On maintaining access to the rooms they’re already in. The influence stays circular: it buys more influence, which buys more access, which buys more influence. Nothing moves.
The test - and really, the key test in 2026 - is whether an institution is willing to spend its Soft Power on a Hard Problem. A problem that affects millions or billions of people. A problem where the cost of action is real and the outcome is uncertain. A shared problem that no single actor can fix alone.
Anthropic recently passed that test. Not perfectly. Not conclusively. Yet visibly, at cost, and in a way that changes the terms for everyone else.
What they did
The mechanics matter, because this is where Soft Power can be evaluated as strategic rather than sentimental.
Saying no was the visible part. The architecture underneath is what made it count:
Their AI Constitution creates written red lines: no assistance with mass-casualty weapons, no enabling of mass surveillance, and no powering of fully autonomous lethal systems.
Their Long-Term Benefit Trust gives a governance body the structural authority to defend those lines over decades, not just over a news cycle.
Their internal Frontier Red Team sits inside the policy division and stress-tests the most advanced systems against exactly these risks.
These go beyond being traditional communications tools. Rather, they’re commitment devices - mechanisms that make the refusal believable. Walking the decision back would require dismantling the architecture - far beyond just changing a press release.
When the Pentagon called, Anthropic had already pre-committed. The Soft Power was structurally embedded.
That’s the difference between Soft Power as reputational nice-to-have and Soft Power as a strategic weapon.
The Hard Problem they’re spending it on
The Hard Problem here isn’t AI safety as an abstract concept. The Hard Problem is two specific system-scale risks: the normalisation of mass domestic surveillance as a standard government capability, and the delegation of lethal force to machines without meaningful human control.
Both risks affect billions of people. Both are entangled with politics, markets, and technology in ways that no single actor can unravel. Both are moving fast enough that the norms being set right now - by labs, by governments, by procurement decisions - will be extremely difficult to reverse in ten years.
Anthropic’s refusal doesn’t solve either problem. More permissive competitors can fill the gap. The Trump administration is already rewarding them for doing so. The substitution risk is real, and anyone saying otherwise is selling a slimmed-down story than the evidence supports.
What the refusal does do: it changes what’s possible to say in every other room where these decisions are being made.
Regulators in Brussels, civil society actors in Seoul, procurement officers in Canberra - they can now point to a major vendor that refused, at scale, at cost. That’s a different kind of proof than a policy paper or a think-tank recommendation. It’s a data point in the global system.
Soft Power spent on a Hard Problem doesn’t fix the problem. It shifts the terms on which the problem gets fought.
Building an index
The Anthropic decision raised a question: who else is doing this? Which institutions - across tech, finance, philanthropy, multilaterals, alliances - are actually spending their Soft Power on Hard Problems, rather than on themselves?
I didn’t find a complete answer. So I started building one, with no-code on Lovable.
The Soft Power Index tracks actors across sectors on three dimensions:
Soft Power: governance quality, willingness to refuse, norm-setting behaviour, and structural credibility.
Hard Problem focus: clarity and seriousness about a specific global challenge: energy transition, critical minerals, AI regimes, supply chain resilience, or democratic institutions.
Capacity: the actual ability to act at scale: relationships, resources, convening power, and reach.
Scores are judgment calls, updated as events unfold. The rubric will evolve with critique. The aim is not to anoint heroes or shame villains. It’s to give leaders a diagnostic: if you want your organisation to matter on the world’s hardest problems, this is one way of asking whether your influence is actually pointed there.
Anthropic sits near the top of the current leaderboard. The Pentagon decision is one of the clearest recent examples of Soft Power spent deliberately, structurally, and at genuine cost on a system-scale risk. The Global South gap in their norm-setting is real. The substitution problem is unresolved. The scores reflect both.
The Soft Power Index is openly maintained. If you have evidence that changes the picture - a decision I’ve missed, a score that’s wrong, an actor that should be on the board - there’s a submission form on every profile. Open data, not open editing.
The question for you:
I built this for leaders whose decisions reach beyond their own organisations or current role. If you’re reading this, you probably already have Soft Power - reputation, relationships, governance, and trust - that you’ve accumulated over years of serious work.
The question the index is designed to force is: what are you spending it on?
Not what panels are you on. Not what access do you have. What Hard Problem - specific, system-scale, and affecting people you will never meet - is your influence pointed at?
Anthropic answered that question under maximum pressure. The answer cost them a nine-figure contract and their relationship with the most powerful government on earth.
That’s one answer. The index exists to find others.
Submit an actor, dispute a score, or tell me who is missing from this initial list - the form is here.
https://softpowerindex.lovable.app
The index is a prototype. It will be wrong about some things. That’s the point - submit your evidence and we’ll update it.
