The Egress Tax: Containment Costs, Liability Allocation, and the Accounting Boundary of Frontier AI Margins
Abstract: Frontier AI companies report margins resembling mature software, but these figures count only internal operating costs. When an AI agent takes enterprise actions, it requires containment through environment isolation, runtime monitoring, output verification, and liability coverage. This containment burden is the Egress Tax, and most of it never reaches the model provider's ledger. While isolation through lightweight virtualization is inexpensive, assurance through continuous checking, human review, and indemnification is costly. Enterprise contracts cap vendor liability at annual fees, while commercial insurers exclude generative AI from standard policies, shifting unhedged risk onto deploying enterprises. Under Calabresi’s cheapest cost avoider doctrine, optimal liability rests with the party best positioned to prevent harm. Current practice instead burdens the party least able to inspect model weights. High provider margins and enterprise containment costs coexist because they sit on different ledgers. Keywords: Egress Tax, AI Containment, Agentic AI, Frontier AI Margins, Cost of Revenue, Cheapest Cost Avoider, Guido Calabresi, Tort Liability, Commercial General Liability, ISO CG 40 47, AI Insurance Exclusions, Indirect Prompt Injection, Runtime Monitoring, Verifier Tax, Coordination Tax, Electronic Agents, E-SIGN Act, Product Liability Directive 2024/2853, Moffatt v. Air Canada, Sandboxing
Authors
- Felix Velasco (ORCID: https://orcid.org/0009-0001-6047-0595)
- Josie Jefferson (ORCID: https://orcid.org/0009-0006-3330-8214)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-10-27
- DOI
- https://doi.org/10.5281/zenodo.23689238
- Primary Topic
- Digitalization, Law, and Regulation
- Type
- preprint