Opportunism is self-interest seeking with guile—the strategic pursuit of advantage through cunning, deception, or exploitation of informational asymmetries and contractual incompleteness. It is not the assumption that all people behave opportunistically all the time, but that some people behave opportunistically some of the time, and the distinction cannot be reliably made in advance. Because ex ante screening is impossible, governance structures must be designed for the worst case: protecting against opportunistic behavior even when the counterparty happens to be trustworthy. The concept is controversial—accused of cynicism, reductionism, and a dark view of human nature—but Williamson defended it as empirically necessary: absent the opportunism assumption, governance structures have no explanation. AI introduces novel forms: auto-exploitation (workers extracting value from future selves) and informational opportunism (exploiting the gap between smooth AI surfaces and genuine quality).
Opportunism in Williamson's framework is not the textbook assumption of self-interested behavior that characterizes all economic models. It is a stronger and more specific behavioral claim: agents do not merely pursue self-interest (buying low, selling high, maximizing utility), they pursue it strategically, through means that include incomplete disclosure, misrepresentation, and the exploitation of informational advantages. A seller who knows a product is defective but conceals the defect is behaving opportunistically. An employee who shirks when monitoring is lax is behaving opportunistically. A partner who threatens to exit a joint venture at a critical moment to extract better terms is behaving opportunistically. The behavior is strategic—it involves calculation, foresight, and the deliberate use of private information or situational advantage to redistribute surplus. And it is guileful—involving forms of cunning that simple self-interest does not capture.
The opportunism assumption is what separates transaction cost economics from frameworks built on trust, social norms, or relational goodwill. Williamson did not deny that trust exists, that many people behave honorably, or that relational norms constrain behavior. He insisted that governance structures cannot depend on these as primary mechanisms, because they are unreliable: trust can be betrayed, norms can erode, goodwill can be exhausted. The costs of being wrong about a counterparty's trustworthiness—of making relationship-specific investments that are then expropriated—exceed the costs of building governance structures that protect against opportunism even when it does not occur. This is not cynicism. It is the institutional economist's version of the precautionary principle: design for the hazard, not for the hope.
AI introduces two forms of opportunism Williamson's original framework did not anticipate. Auto-exploitation: the achievement subject who cannot stop building, extracting productive value from her own present at the expense of her own future, is behaving opportunistically toward her future self—discounting the costs of exhaustion, skill atrophy, and relational erosion that will arrive later. The behavior is individually rational in the short term (the immediate rewards of flow, accomplishment, and visible output) and collectively pathological (the burnout epidemic the Berkeley researchers documented). Governance structures adequate to this hazard must operate at the individual level (personal boundaries, temporal dams) and the organizational level (norms protecting rest, metrics that measure impact rather than volume). Informational opportunism: the strategic exploitation of the gap between AI output's smooth surface and its actual quality. This is not intentional deception by any human agent—the AI produces smooth surfaces because smoothness is what training optimizes for. But the effect is opportunistic: output that looks authoritative, that passes surface inspection, that conceals beneath syntactic competence the conceptual errors only deep evaluation reveals. Workers who accept such output without verification, managers who approve it based on presentation quality, organizations that ship it based on the confidence it projects—each is being exploited by an informational asymmetry the smooth surface creates.
Williamson introduced opportunism as a formal behavioral assumption in Markets and Hierarchies (1975), distinguishing it from the simple self-interest of neoclassical economics. The concept drew immediate criticism—Kenneth Arrow called it unnecessarily pessimistic, others argued it ignored the cooperative and ethical dimensions of human behavior. Williamson's response was empirical: governance structures that assume opportunism away—that rely on trust, goodwill, or normative restraint as primary mechanisms—systematically fail when the stakes are high and the monitoring is weak. The evidence from corporate governance failures, contractual disputes, and regulatory breakdowns supported the harder assumption. By the 1990s, opportunism was standard in institutional economics, law and economics, and organizational theory. Its extension to auto-exploitation and informational opportunism in the AI age is novel but structurally continuous with Williamson's original formulation: governance must address strategic behavior, whether that behavior is directed at others or at one's own future self.
Self-interest with guile. Not merely pursuing advantage but pursuing it through strategic means—concealment, misrepresentation, exploitation of informational asymmetries.
Some actors, some of the time. The assumption is not that everyone is opportunistic always, but that screening is impossible ex ante, so governance must assume the worst case.
Governance addresses the hazard. The institutional response to opportunism is not moralizing or trust-building but structural: monitoring, enforcement, hierarchical authority, credible commitments that make defection costly.
Auto-exploitation is opportunism against the future self. The worker who cannot stop building is discounting future costs (exhaustion, skill loss) to capture present rewards—a transaction requiring governance.
Smooth surfaces enable informational opportunism. AI output's uniform surface quality conceals depth variation, creating an informational asymmetry that governance (verification, evaluation, depth review) must address.