Regulation
California's AB 316, in force since 1 January 2026, bars any business that developed, modified or deployed an AI system from arguing the AI "acted autonomously" to escape liability, a defence collapse that lands squarely on firms operating money-moving trading agents
The law does not create new strict liability for AI harm, it removes exactly one argument, that autonomous behaviour breaks the chain of responsibility, while leaving every other defence intact. For anyone deploying an agent that places trades, the practical read is that every transaction is now attributed to a person, and the plaintiff gets to pick which one.
AB 316 is narrow by design, and that narrowness is what makes it consequential for trading agents specifically. The statute adds one clause to California's civil code, that a defendant in an AI-harm case cannot claim the system acted autonomously as a way of severing the causal chain back to a human or corporate actor. It does not establish strict liability, does not ban autonomous trading, and leaves causation, foreseeability and comparative-fault defences fully available. What it removes is the single argument that agentic AI's own advocates have leaned on hardest, that a sufficiently autonomous system's decisions are not fairly attributable to whoever built or deployed it.
The law's reach runs the length of the supply chain, not just the end user. Foundation model developer, the firm that fine-tunes or customises the model for trading, the platform that integrates it into an execution pipeline, and the brokerage or fund that deploys it against live capital are all captured by the same "developed, modified, or used" language. For the trading-agent stack specifically, layered as it typically is across a model provider, a strategy or signal vendor, and a brokerage execution layer, that means a single bad trade can now draw a liability claim against multiple parties at once, each unable to point at the model and call it the responsible party.
This lands at a specific moment for the industry. Robinhood, Public and a growing list of brokerages have spent 2026 rolling out agentic trading products that let third-party AI agents place orders with limited or no per-transaction human approval, precisely the deployment pattern AB 316 is aimed at. The legal guidance emerging around the law is blunt about what that means in practice, treat every transaction an agent executes as attributable to a specific person, and assume the plaintiff's counsel gets to choose which person in the chain that is.
For firms building or operating these systems, the practical response the law is already pushing is architectural rather than purely legal, human-approval defaults on consequential transactions, tightly scoped agent permissions, hard spending caps, and transaction logging sufficient to reconstruct exactly which human decision authorised which trade. California is one state, but AI liability law tends to travel, and a brokerage building its agentic trading stack to survive AB 316 scrutiny in California is, in practice, building the version that survives scrutiny everywhere else too.
Read the original: Astraea Counsel - California's AB 316, in force since 1 January 2026, bars any business that developed, modified or deployed an AI system from arguing the AI "acted autonomously" to escape liability, a defence collapse that lands squarely on firms operating money-moving trading agents. Commentary is the independent editorial view of Share Trading; the original article is credited to its publisher.