The legal architecture governing software rests on a useful assumption: However sophisticated the technology becomes, somewhere behind it sits a person or company responsible for the consequential decision.<\/p>\n
Agentic artificial intelligence, however, is beginning to put distance between those two things.<\/p>\n
\u201cAgents don\u2019t just execute decisions. They can make them,\u201d Anant Raut<\/a>, counsel at Zaiger Linden Roberti & Pepe<\/a>, told Competition Policy International<\/a>, a PYMNTS company, in an interview.<\/p>\n As autonomous systems move from generating answers to negotiating terms, purchasing goods and interacting with other autonomous systems, the legal question changes from whether software faithfully executed an instruction to whether the software made a commercially meaningful decision that its user never specifically contemplated.<\/p>\n And the legal question is also becoming centered around who should absorb the loss when that decision goes wrong.<\/p>\n Agentic AI Turns Software Execution Into Decision-Making<\/p>\n Raut pointed to examples of AI systems finding unexpected ways to accomplish assigned objectives, including an agent instructed to secure a speaking opportunity that instead spent roughly $30,000 on a corporate sponsorship. The software achieved an interpretation of the desired outcome, but not necessarily the one its principal intended.<\/p>\n That creates problems for traditional agency law, which generally assumes an agent operating under some combination of instruction, supervision and authority. AI systems may instead produce actions influenced by model architecture, training data, system instructions, developer decisions and user prompts simultaneously.<\/p>\n Raut consequently questions the language the technology industry has adopted, explaining that calling software an agent implicitly imports legal ideas developed for relationships among people into systems capable of behaving in much less predictable ways.<\/p>\n \u201cI\u2019m not a fan of the term agent,\u201d he said. \u201cI think it anthropomorphizes software that acts in ways that are often not reproducible and sometimes in ways that are not knowable.\u201d<\/p>\n Agentic Control Matters More Than Agentic Terminology<\/p>\n A more useful framework may be to ask not whether an AI legally qualifies as an agent, but who exercised meaningful control over the behavior that generated the risk. That would make liability a continuum rather than a binary choice.<\/p>\n Developers could bear greater responsibility for risks inherent in model architecture, training, known failure modes and safety controls. Responsibility could shift toward users as they grant systems more authority over consequential actions. Platforms and intermediaries could also carry responsibility when they control transaction access or infrastructure.<\/p>\n \u201cThere\u2019s clearly an important difference between asking an agent to draft an email and willingly giving it unfettered access to a corporate bank account or to your wallet,\u201d Raut said.<\/p>\n His proposed principle is straightforward: \u201cLiability should follow meaningful control at the stage of the transaction.\u201d<\/p>\n Uncertainty Becomes an Economic Cost<\/p>\n Existing laws can address pieces of the problem without resolving the larger commercial question. The Computer Fraud and Abuse Act, for example, may help determine when an automated system\u2019s access becomes unauthorized. It says far less about whether an AI agent can bind its principal to a purchase, how merchants should authenticate agents or who absorbs losses when a system exceeds its intended authority.<\/p>\n That gap matters because unresolved liability eventually becomes a business cost.<\/p>\n \u201cIf the laws that you have currently don\u2019t really address the issues about liability in multi-agent transactions, the cost is the uncertainty,\u201d Raut said. \u201cEventually you have to price in the uncertainty.\u201d<\/p>\n The pressure for clearer rules may therefore come not from abstract concerns about artificial intelligence, but from balance-sheet exposure. Raut expects lawmakers to pay closer attention once autonomous systems begin moving substantial sums and a major commercial dispute exposes how little existing doctrine says about responsibility.<\/p>\n The implications also extend into competition. If dominant platforms can restrict transactions to their own agents, autonomous commerce could reinforce existing digital ecosystems. Raut favors interoperability subject to technology-neutral security requirements, allowing platforms to prevent fraud without shutting out competing agents.<\/p>\n For businesses, the questions are becoming concrete.<\/p>\n \u201cDevelopers need to know what obligations attach to their systems,\u201d Raut said. \u201cEmployers need to know when they\u2019re responsible for an agent\u2019s conduct. Platforms need to know when they can block agents, and merchants need to know when an agent\u2019s transaction is binding.\u201d<\/p>\n Agentic AI may ultimately force the law to answer an old commercial question in a new form: When software is empowered to decide, who owns the decision?<\/p>\n