AI Layoff Trap

The game-theoretic case that an AI mass-layoff wave could happen even though every CEO sees it coming and wants to avoid it. Named by Gerry Tsoukalas & Brett Falk's paper "The AI Layoff Trap" (captured via The AI Layoff Trap (BBC Global)). Workers are also customers; firing them destroys demand — but in a competitive market each firm bears almost none of the demand it destroys, so replacement becomes a dominating strategy and the economy walks into a prisoner's dilemma where "demand is entirely eroded and every firm is worse off."

The mechanism

Monopoly internalizes, competition externalizes. A monopolist firing 90% of its staff fires its own customers, so it self-restrains. A firm among hundreds reasons "my workers only spent a fraction with me anyway" — the demand cost rounds to ~zero while the cost savings are certain.
Defection dominates. If rivals automate and you don't, you're bankrupted and your workers lose their jobs anyway. Whatever others do, maximal automation is your best reply — the definition of a dominant strategy.
Agreements are unstable. Any voluntary slow-down pact (the paper models even an Anthropic/OpenAI/Microsoft/Google handshake) collapses on exit from the room: defecting is both maximally profitable and arguably a fiduciary duty. The paper's formal prediction: firms will not self-restrain.

The policy answer the model selects

Of ~six instruments tested (UBI, worker equity, etc.), only one works: directly repricing the replace-vs-augment margin —

a tax on fully replacing a worker with AI (augmentation stays costless) — carbon-tax-style in design;
or its politically softer twin, subsidies for retaining workers.

The design principle is the load-bearing idea: don't tax AI adoption, tax substitution. That draws exactly the augment/replace line that the current empirical evidence (Skill-Biased Labor Augmentation) says the market is — for now — drawing voluntarily.

Status: conditional model, not current observation

The trap activates only if/when AI can fully substitute for large job categories. The vault's evidence stack says that condition isn't binding yet:

AI Adoption and Headcount Growth — adopting firms are growing headcount, entry-level fastest
Skill-Biased Labor Augmentation — McCrory: no material AI unemployment effect to date; persistent returns to expertise
Downside scenario if it does bind (Tsoukalas): unprecedented wealth inequality → political instability

Hold both without contradiction: one models the bad equilibrium, the other measures the current good one, and the hinge variable — augment vs replace — is the same in both. Whittemore's self-fulfilling-narrative point (see Why AI Hasnt Increased Unemployment According to Anthropic (AI Daily Brief)) is the bridge: which equilibrium we land in is partly a function of which story executives act on.

Connects to your work

For an enterprise IT leader, the trap is the strongest steelman of the pessimist case to carry into leadership conversations — it doesn't require bad actors, just ordinary competitive rationality. Practical uses: (a) it upgrades "augment, don't replace" from a values statement to an economy-level stability argument; (b) it's the intellectual counterweight to pair with the augmentation data so optimism doesn't read as naivety; (c) the replace-vs-augment margin is exactly the design choice in every agentic-transformation program (cf. Agentic Pods — augmentation operationalized).

Cross-references

The AI Layoff Trap (BBC Global) — anchoring source
Skill-Biased Labor Augmentation · AI Adoption and Headcount Growth — the current-evidence counterweight
FOBO (Fear of Becoming Obsolete) — the trap is the macro-scenario that would vindicate the anxiety
Cobra Effect — kindred incentive-design pathology
AI Wealth Fund — redistribution proposals the model finds insufficient (they don't touch the incentive)
Jevons Paradox — the opposing scope-expansion mechanism