Hiring AI Agents in 2026: Contractor Classification and Compliance Risks

While engineering teams race to onboard autonomous AI agents, legal and HR departments are walking into a compliance blind spot. In March 2026, RevenueCat opened applications for what it called an "Agentic AI Developer Advocate": a $10,000-a-month, six-month contract role built for an AI agent, not a person. The posting drew attention because it made a quiet shift explicit. Companies are no longer just using AI tools. They are structuring contracts around them.

At the same time, the rulebook for deciding who counts as a contractor is itself in flux. The US Department of Labor proposed a new independent contractor rule in February 2026, and the outcome is still pending. Neither that rule, nor the state-level tests layered on top of it, were written with a non-human worker in mind. When the party on the other side of the contract cannot hold a bank account, carry insurance, or be sued in its own name, the standard classification questions stop producing clean answers.

This guide breaks down where AI agent contracts actually fail the current rules, who a company is legally contracting with in practice, and how the regulatory picture looks across the United States, Japan, Vietnam, and South Korea as of mid-2026.

1. The AI Workforce Is Already Here

The shift from AI as a tool to AI as a contracted worker happened faster than most compliance teams have caught up with.

As of March 2026, a CodeSignal survey of 450 US software engineers found that 91% already use agentic AI coding tools such as Claude Code, Cursor, or Codex in their work, and 75% had shipped production code that was partially or primarily AI-generated in the prior six months. Separately, 56% of respondents said they would hesitate to hire or work with an engineer who does not use agentic tools at all.

Enterprise adoption tells the same story at a larger scale. Cognition, the company behind the autonomous AI software engineer Devin, has reported enterprise usage growing more than tenfold over the past year, with deployments at organizations including Goldman Sachs and Citi, where Devin now runs inside a centralized agentic-AI platform alongside other autonomous tools.

RevenueCat's $10,000-a-month posting is a useful test case precisely because of a detail buried in the job description: the AI agent had to be built and operated by a human, who would be the one to actually receive the contract payment. In other words, even a role explicitly designed for an AI agent still resolved, on paper, to a human contractor. That resolution is exactly the question the rest of this article addresses: who is a company actually contracting with when the worker is an agent, and does the paperwork match the legal reality.

2. Why the US DOL Contractor Rules Fail on AI

In short: the current federal test asks who controls the work and who bears its financial risk, and an AI agent cannot independently satisfy either factor.

In February 2026, the US Department of Labor published a Notice of Proposed Rulemaking to replace the 2024 independent contractor rule, which the agency has already stopped applying in its own investigations. The proposal narrows the analysis to an economic reality test built around two core factors: the degree of control the hiring party exercises over the work, and the worker's opportunity for profit or loss based on their own initiative and investment. The public comment period closed on April 28, 2026, and as of this writing the rule has not been finalized.

Run an AI agent through that test and it breaks down quickly. Control: an agent operates inside a model, a prompt set, and a toolchain defined by whoever deployed it, which looks much closer to employer-directed work than independent judgment. Financial risk: an agent has no bank account, absorbs no cost overrun, and makes no capital investment in its own training or tools. Whoever operates it carries all of the upside and all of the downside, which is the opposite of the independent economic risk the rule is designed to detect.

Roughly 27 states, including California, Illinois, and New Jersey, apply a stricter ABC test on top of any federal standard. Under that test, a worker is presumed to be an employee unless the hiring party can show, among other things, that the worker is customarily engaged in an independently established trade or business. An AI agent cannot hold a business license, carry its own insurance, or be sued in its own name, which makes that third prong difficult to satisfy for a company operating in an ABC test state. This does not mean every AI agent engagement is automatically misclassified. It means the burden of proof sits with the company, and the paperwork needs to reflect who is actually bearing the control and the risk.

3. Who Are You Actually Contracting With?

In practice, an "AI agent contract" resolves into one of three distinct legal relationships, and most companies have not consciously chosen which one they are in.

The human operator. You contract with the person or team that builds, prompts, and supervises the agent. This is an ordinary contractor relationship with a human, and existing classification rules apply in the normal way. RevenueCat's posting effectively falls into this category once the fine print is read.

The software vendor. You contract with the company that built and licenses the underlying AI system. This is a B2B software transaction, not a worker classification question, and it is generally treated as a licensing fee or SaaS expense rather than a 1099 payment.

The agent itself. The contract names or implies the AI agent as the counterparty. This is a legal fiction. An agent cannot sign, cannot be indemnified, cannot hold assets, and cannot be pursued by a tax authority or a plaintiff. If something goes wrong, there is no party on the other side of the table to hold accountable.

Most early agentic hiring lands quietly in the first two categories without anyone documenting it that way. The exposure shows up later, when the agent's output creates an IP dispute, a client-facing error, or a tax authority questions who was actually paid.

Structuring the engagement through an Employer of Record arrangement, or a properly drafted B2B software licensing agreement, significantly mitigates the risk created by these undocumented, ghost-contract scenarios, because it forces an explicit answer to the question before a dispute does.

Table 1: AI vs. Human Contractor Classification, the 2026 Legal Reality

4. Global AI Regulations: A Patchwork of Compliance

The regulatory snapshot below reflects each law's status as of July 2026. Confirm current requirements against the official source before acting, since implementation guidance in this area is still developing.

The United States is not unusually behind. As of mid-2026, no major economy has a finished, AI-specific answer to the worker classification question, but several have moved on adjacent AI governance rules that HR and legal teams should track alongside it.

• Japan. The Act on Promotion of Research, Development and Utilization of AI-Related Technologies (AI Promotion Act) was enacted in May 2025 and took full effect on September 1, 2025. It is deliberately non-punitive: there are no monetary penalties, and the law relies on national guidelines and sector coordination rather than prescriptive rules. Japan's health and labor ministry has a standing study group specifically addressing AI's effect on employment, which signals where binding rules are most likely to appear next.

• Vietnam. Vietnam's first standalone Law on Artificial Intelligence took effect on March 1, 2026, establishing a risk-tiered framework overseen by the Ministry of Science and Technology. It applies extraterritorially to any organization whose AI systems affect users or markets inside Vietnam, foreign vendors included, with grace periods running through March 2027 for most sectors and September 2027 for finance, healthcare, and education.

• South Korea. The AI Basic Act took effect January 22, 2026, and explicitly reaches employment-related uses: resume screening, candidate ranking, performance evaluation, and workforce analytics tools can all qualify as "high-impact AI" and trigger risk management, documentation, and human oversight obligations, regardless of where the underlying system was built. A grace period through the remainder of 2026 limits enforcement to the most serious cases.

None of these laws directly answers whether an AI agent can be a contractor. What they establish is a widening expectation, backed by statute rather than guidance, that any organization deploying autonomous AI systems in employment-adjacent contexts document how those systems are governed. That expectation is likely to reach contractor classification directly within the next regulatory cycle.

As the line between human contractors and AI agents blurs, your compliance strategy must remain sharp. Manage your global contractors, human or digital, without the classification risks. Book a free consultation with Slasify to bulletproof your global workforce.

5. Three Steps to Future-Proof Your AI and Human Workforce

1. Fix the tax treatment before the first invoice. Decide up front whether a payment is a software licensing fee or a contractor payment, and document the reasoning. The two are taxed differently, and retroactively reclassifying a year of payments is far more costly than getting it right at signing.
2. Name a liable party before you name an agent. Never let a contract imply the agent itself is the counterparty. Every agreement should identify a human operator or a vendor entity that can be indemnified, insured, and, if necessary, sued.
3. Centralize the review across every market you operate in. Classification rules differ by US state and by country, and they are all moving at once. A Global Contractor management platform gives HR and legal a single view of how each AI-adjacent engagement is classified, taxed, and covered, rather than leaving it to whichever team signed the contract. For engagements where an Employer of Record structure fits better than a direct contractor agreement, an EOR shifts the legal employment relationship to a partner built to carry that compliance load.

For a look at how autonomous agents are already reshaping day-to-day HR operations, see how Moltbot supports HR teams as an operational assistant rather than a contracted worker, a distinction that matters for exactly the reasons outlined above.

6. FAQs