The experiment was called OpenClaw. A group of biohackers attempted to perform neurosurgery on a live lobster, wire its peripheral nervous system to an external interface, and then cede motor control to an agentic AI system. By most accounts, the results were both predictable and grim. But before this gets filed away as a curiosity about the outer edges of internet culture, it is worth asking a harder question: under what regulatory authority, in Canada or anywhere else, would this experiment have been clearly, unambiguously illegal?
The honest answer is: it depends, and that dependence is itself the problem.
What Actually Happened
The project, documented and publicized online, involved biohackers attempting surgical intervention on a crustacean to externalize and intercept neural signals. The stated goal was to feed those signals into an AI model — OpenClaw — which would then interpret and issue motor commands back to the animal, effectively substituting artificial decision-making for the lobster’s own nervous system responses.
Whether the experiment achieved anything resembling its stated goal is secondary to what it represents: an uncontrolled, non-institutionally sanctioned attempt to create a closed-loop system between a living organism’s nervous system and an agentic AI. No ethics board. No biosafety review. No institutional oversight of any kind.
This is not a hypothetical edge case that regulators might eventually need to consider. It happened. And it will happen again, with more capable systems and more complex organisms.
The Regulatory Blind Spot
Canada has developed increasingly serious AI governance infrastructure over the past several years. The Artificial Intelligence and Data Act, embedded within the now-stalled Bill C-27, attempts to establish a risk-based framework for high-impact AI systems. The federal government has published voluntary codes of conduct for generative AI developers. The Office of the Privacy Commissioner has issued guidance on AI and personal data.
None of this touches what OpenClaw represents.
Canada’s animal welfare protections — governed primarily at the provincial level, with federal jurisdiction applying mainly to agriculture and transport — were not designed with neurosurgical AI integration experiments in mind. The Criminal Code provisions around animal cruelty require proof of wilful neglect or unnecessary suffering, standards that become conceptually murky when the agent causing the suffering is an AI system acting on signals it was given control over.
Meanwhile, the bioethics frameworks that govern human research — the Tri-Council Policy Statement, institutional review boards, Health Canada’s clinical trial regulations — explicitly exclude non-human animals from their scope in most contexts, deferring instead to the Canadian Council on Animal Care. The CCAC sets standards for animals used in science, but its authority applies to institutions. Independent biohackers operating outside any university, hospital, or corporate research environment occupy a gap the framework was never designed to close.
When you layer agentic AI on top of that gap, the ambiguity compounds. Is OpenClaw a device? A software system? A research tool? Who bears liability for the decisions it makes once given control of a living nervous system — the developers, the biohackers, the platform that hosted the model?
Agentic AI Changes the Stakes
The OpenClaw experiment matters beyond its shock value precisely because of the agentic dimension. This was not a researcher using AI to analyze pre-recorded neural data. The design intent was real-time, closed-loop control: the AI system receiving biological signals and issuing commands back to a living body. That is a meaningfully different category of interaction, and it is one that existing AI governance frameworks have not seriously grappled with.
Agentic AI — systems that pursue goals, take sequential actions, and interact with external environments autonomously — is advancing rapidly. The same architectural principles behind customer service agents and code-writing copilots can, in principle, be applied to biological interfaces. As brain-computer interface research matures, as biosensor technology becomes cheaper, and as open-source AI models proliferate, the barrier to conducting exactly these kinds of experiments will continue to fall.
Canada’s current approach to AI governance, risk-based and largely focused on data processing and automated decision-making in commercial contexts, has no conceptual vocabulary for this. The risk taxonomy in AIDA does not include a category for AI systems that interface directly with living nervous systems outside clinical settings.
What a Serious Policy Response Looks Like
Addressing this gap does not require panicked legislation. It requires deliberate expansion of existing frameworks to acknowledge that the boundary between digital AI systems and biological systems is no longer theoretical.
- The Canadian Council on Animal Care’s mandate should be explicitly extended to cover independent and community-based research contexts, not only institutional ones, with clear guidance on AI-assisted or AI-directed animal experimentation.
- AIDA’s risk classification system, if and when it becomes law, should include explicit provisions for AI systems that interact with biological organisms in real time — animal or human — outside regulated clinical or research environments.
- Health Canada and the Public Health Agency of Canada should jointly assess whether biosafety and bioethics review mechanisms need an independent-researcher carve-in, particularly as synthetic biology and biohacking communities grow in sophistication.
- Provincial attorneys general should clarify, in guidance if not in statute, how animal cruelty provisions apply when harm is mediated through an automated or autonomous system.
The OpenClaw experiment will likely be remembered, if it is remembered at all, as an embarrassing stunt. But embarrassing stunts have a way of becoming precedents. The serious question is not whether the biohackers involved understood what they were doing — clearly, they did not, fully. The serious question is whether Canada’s regulatory architecture is capable of recognizing this category of risk before it scales into something that cannot be addressed after the fact.
Right now, the answer is no.
Related InsightTrack Analysis
- The Sovereignty Gap: Why Canada’s Agentic AI Boom Depends on Foreign Model Providers
- Canada’s Sovereign AI Push Meets the Chip Wall
- Agent Orchestration Is Reshaping How AI Actually Gets Work Done
