The captivity assumption is breaking across platforms, states, and enterprise software — exit optionality is appreciating faster than the coercive architectures priced to ignore it.
The Pattern
## The Pattern
Under Secretary of War for Policy Elbridge Colby summoned Pope Leo XIV's ambassador to the Pentagon in January. In a closed-door lecture about American military supremacy, a US official invoked the Avignon Papacy, the fourteenth-century episode when the French Crown used military force to bend the bishop of Rome.
The Cardinal sat through the lecture in silence. The Holy See has not, since that day, given an inch. The Vatican declined the 250th anniversary visit. Pope Leo escalated his public posture rather than softening it. The hard-power actor discovered the soft-power institution had no capture point to exploit.
Now read that pattern in four other places today.
CyberAgent reached 93% monthly active usage of ChatGPT Enterprise without mandating it. The company explicitly declined to enforce adoption. Each team evaluated the tool against alternatives. Adoption spread through culture-building, shared prompts, usage rankings, and Slack bots that asked non-users what they needed instead. The Japanese conglomerate got enterprise penetration by removing the captive-audience architecture, not installing it.
A developer publicly reallocated $100 monthly of Claude Code spend toward Zed and OpenRouter. The cited reason was not model quality. It was that OpenRouter credits expire after 365 days, so the user captures unused time instead of forfeiting it to the platform. The rebellion is accounting.
The EFF announced it was leaving X after almost twenty years. A single 2018 tweet generated more impressions than an entire year of current posts. The decision was arithmetic. Staying cost more than leaving, measured honestly.
And Cloudflare previewed EmDash, a TypeScript CMS whose core architectural claim is sandboxed plugins. The pitch cites that 96% of WordPress vulnerabilities come from plugins with full file and database access. The infrastructure assumes the extension ecosystem is hostile by default.
Coercion failing against the Vatican. Voluntary adoption reaching 93%. Customers rerouting around usage caps. Institutions leaving platforms on spreadsheet grounds. Infrastructure rebuilt to sandbox its own extensions. Different domains. Same observation: the captivity assumption is breaking everywhere at once.
The Tension
## The Tension
Two incompatible operating theories are in production right now.
Theory A says users and institutions are captured by switching costs, network effects, legacy integrations, or political exposure. You can extract from them because they cannot credibly leave. Most enterprise SaaS pricing still prices on Theory A. Most platform ad inventory still prices on Theory A. Most state-power calculations about leverage over transnational institutions still price on Theory A.
Theory B says the cost of leaving collapsed quietly. OpenRouter rotates models in an afternoon. Git migrates any CMS. Bluesky absorbs fleeing institutional accounts. The Vatican does not need the United States for its 250th anniversary. It needs Lampedusa, and it chose Lampedusa.
The tension for builders is that Theory A still looks profitable on this quarter's dashboard. Microsoft's OneDrive dark patterns still generate subscription conversions from users who do not understand what was silently redirected. X still collects advertising revenue. Claude's fixed-plan pricing still extracts from users who resent it.
But Theory B is eating the foundations underneath the cash flows. Every team has to decide which theory to build for. The answer is not obvious if you are reading engagement metrics instead of exit velocity.
What This Unlocks
## What This Unlocks
Build assuming your users can and will leave. This is an architecture question, not a philosophy one.
If you are building infrastructure, EmDash's approach is now the shape of the thing. Sandboxed plugins. Programmable access. Pay-per-use primitives like x402. The CMS that assumes its plugin ecosystem is hostile ships faster than the one that audits every plugin for trust.
If you are building enterprise software, CyberAgent's adoption model is the reference case. You do not drive 93% usage by mandating it. You drive it by building a culture where using the tool is socially reinforced and alternatives are evaluated honestly. The company that skips the mandate gets more adoption than the company that enforces it. That is the structural finding.
If you are building pricing, the OpenRouter-versus-subscription story tells you what to ship. Credits that accrue. Rolling windows. Granular billing. Loss aversion to forfeited credits is now a larger force than vendor gain from unused subscription months.
Stop building captive-audience funnels, default opt-ins for paid storage, and architectures that make leaving expensive. Not because they are unethical, though they are. Because they are now detectable, and one blog post can expose the funnel to everyone at once.
Watching Next
## Watching Next
1. Whether Anthropic ships rollover credits for Claude plans within 60 days. The defection math is public, the customer cited it by name, and the pressure is already priced into the discourse. 2. Whether three or more peer NGOs follow EFF off X within 30 days. EFF's math applies to every organization still posting there; publication accelerates coordination. 3. Whether the Vatican extends its public posture into direct economic or diplomatic pressure on US Iran operations within April. Leo's statements have escalated weekly since January. 4. Whether Cloudflare's x402 primitive gets adopted by at least one non-Cloudflare CMS within 90 days. If yes, pay-per-crawl becomes the default AI-agent access pattern. If no, it stays a Cloudflare lock-in play.
Underweighting
## What the Market Is Underweighting
Everyone is pricing AI model capability. Almost no one is pricing exit optionality.
The developer who reallocated his Claude spend did not switch because a different model was better. He switched because his credits expired on a schedule he did not control. That is a pricing architecture failure, not a capability failure. The market is still trading model benchmarks as the scarce variable when the scarce variable is actually control over the billing window.
The Vatican story is also mispriced. Most observers will read it as a one-off diplomatic incident. It is actually a clean test of whether the coercion toolkit still works on institutions with long time horizons. The Holy See thinks in centuries. The Pentagon thinks in news cycles. When those time horizons collide, the longer horizon wins — and the precedent propagates to every other long-horizon actor watching.
The thing being underweighted is not a technology trend. It is a reprice. The discount rate on exit has fallen, and assets priced on captive-audience assumptions are quietly becoming worth less. Enterprise subscription software. Platform ad revenue. State-power leverage over transnational institutions. Default-setting dark patterns. All of these are marked to a model that assumed the user could not leave. The model is wrong now, and the correction has not hit the tape.
Bottom Line
## Bottom Line
Coercion is losing to optionality in real time. If you are building anything whose revenue depends on users not leaving, audit the exit cost today — your moat is probably repricing faster than your dashboard shows.
Sources
34 articles scanned / 5 sourcesGet The Signal daily
Cross-domain structural analysis, delivered every morning.