What Actually Happened

Getty Images and Shutterstock announced a $3.7 billion merger agreement that cleared the U.S. Department of Justice but was effectively killed by the UK's Competition and Markets Authority, which imposed restrictions that would have excluded a material portion of Shutterstock's business from the combined entity. Getty has since moved to terminate the deal. This is not a story about antitrust law per se. It is a story about what happens when two regulatory jurisdictions operate with structurally incompatible schemas about what a market is and what harm looks like inside it.

Two Jurisdictions, Two Structural Models

The DOJ and the CMA looked at the same merger and reached different conclusions. That divergence is not simply a matter of stricter versus looser enforcement. It reflects something more fundamental: each regulatory body is working from a different structural model of the stock image market, including different assumptions about competitive harm, market definition, and what counts as a protected asset. The DOJ appears to have accepted that the merger would not substantially lessen competition in its relevant markets. The CMA drew a different boundary around what the "market" includes and what portions of that market deserve protection.

This is exactly the distinction Gentner (1983) draws between surface similarity and structural similarity in analogical reasoning. The two regulators share surface features - both are antitrust authorities reviewing the same transaction - but they are operating from different relational schemas about how image licensing markets are organized, who the relevant competitors are, and which structural features of the deal generate harm. They matched on objects but not on relations.

Why This Matters for the Current AI Governance Debate

The Forbes report on AI governance published this week notes that Anthropic, OpenAI, the Vatican, and Congress all agree AI needs guardrails but disagree on what should be protected first - catastrophic risk, human dignity, or U.S. competitiveness. This disagreement is being treated primarily as a values conflict. I think that framing is partially wrong. The disagreement is also a schema conflict, and schema conflicts produce outcomes like the Getty-Shutterstock collapse: regulatory fragmentation that destroys value without producing coherent protection.

The awareness-capability gap that runs through my dissertation research applies here in a non-obvious way. Regulators across jurisdictions are clearly aware that AI systems pose coordination problems. But awareness of the problem does not translate into compatible governance responses. The CMA and the DOJ were both aware that stock image market concentration was a concern. That shared awareness did not produce compatible structural models of the problem, and the result was a deal that survived one jurisdiction's scrutiny and failed another's on what appear to be irreconcilable definitional grounds.

The Folk Theory Problem in Regulatory Design

Kellogg, Valentine, and Christin (2020) document how workers develop folk theories of algorithmic systems - idiosyncratic, experience-based impressions that are not structurally grounded and therefore do not generalize well. Regulatory bodies face an analogous problem with novel markets. When a market structure is genuinely new, whether it is AI-generated content licensing or large language model API access, regulators tend to build folk theories by analogy to prior markets they understand. The CMA may be reasoning from its experience with media consolidation in the UK press market. The DOJ may be reasoning from its framework for software platform markets. Neither analogy maps cleanly onto the structural features of AI training data and synthetic image generation, which is where this market is actually heading.

Hatano and Inagaki (1986) distinguish routine expertise, which is optimized for familiar task structures, from adaptive expertise, which involves understanding deep principles well enough to respond to structurally novel conditions. The Getty-Shutterstock outcome is an object lesson in what happens when routine regulatory expertise encounters a market in structural transition. The procedures are sound. The schema is outdated.

The Practical Implication for AI Governance Architecture

If Anthropic, OpenAI, and Congressional staff are currently in the process of building the governance frameworks that will apply to AI systems over the next decade, the Getty-Shutterstock collapse is a concrete warning about what jurisdictional schema fragmentation costs. The deal was not small. The regulatory divergence was not about procedure. It was about what structural features of the market each authority decided to make central to its analysis.

The organizations currently negotiating AI governance frameworks would benefit from treating schema alignment as a prior task rather than an assumed condition. Before agreeing on rules, jurisdictions need to agree on the relational structure of the thing being governed - what the relevant units are, what harm relations connect them, and which structural features are load-bearing for the analysis. Without that alignment, the result is governance that clears one jurisdiction and fails another, which for globally deployed AI systems is not a minor inconvenience but a structural barrier to any coherent coordination at all.