AI’s benefits may justify taking risks. But who decides which risks are acceptable - and who pays when they materialize?
In a Politico interview reported by Reuters on October 4, OpenAI CEO Sam Altman argued for broad AI access despite some harmful outcomes. His examples included hacks, scams and misuse. He described “a lot of daylight” between OpenAI and Anthropic’s regulatory outlooks and defended a lighter-touch approach. The remarks express a policy position; they do not announce a new law or formal company policy change. Reuters
For Canadian lawyers, the significance lies in the distinction between accepting risk as a matter of public policy and allocating responsibility for a particular loss.
The Benefits and the Losses May Reach Different People
A technology can deliver widespread benefits while causing serious harm to individuals or businesses.
Consider a hypothetical AI-enabled scam. Thousands of users may benefit from the same technology’s legitimate applications. The person who loses their savings nevertheless faces a specific injury, with questions about prevention, evidence and recovery.
That is the analytical problem behind the debate. An assessment of overall social benefit does not establish what precautions a particular developer or deployer should take, or resolve who should bear a particular loss.
Nor should Altman’s examples be expanded into a claim that he endorsed catastrophic risk. The reported discussion concerned access, misuse and the trade-offs associated with his preferred regulatory approach.
“Acceptable Risk” Needs a Defined Context
For corporate counsel, a broad statement about tolerating risk becomes useful only when translated into a specific decision.
Which system is being deployed? Who will use it? What could go wrong? How severe would the consequences be? Which safeguards are available?
A drafting assistant used by trained employees presents different questions from an autonomous tool permitted to communicate with customers, transfer funds or modify systems.
The relevant assessment should therefore identify the proposed use, the affected people and the limits on the tool’s authority. A company’s general enthusiasm for AI cannot substitute for that analysis.
What Canadian Lawyers Should Examine
The following are practical implications of the policy debate, rather than conclusions about OpenAI’s legal liability.
For litigation counsel, a dispute involving AI-related harm may require close attention to what participants knew, what they could control and how the alleged loss occurred. A public statement can supply context, but its legal significance would depend on the claim, evidence and applicable law.
For corporate counsel, the priority is documenting deployment decisions. Who reviewed the risks? What restrictions were imposed? What evidence would trigger a reassessment?
For procurement counsel, the issue is how the contract distributes responsibilities. Buyers should understand the permitted uses, security commitments, incident-notification provisions, liability limits and available remedies.
These questions connect the public argument about AI access with the decisions organizations make when adopting a product.
Incident Reporting Makes Risk Visible
A discussion about tolerable harm also needs a way to identify harm when it occurs.
An internal reporting process should distinguish an actual incident from a near miss, assign responsibility for investigation and preserve relevant records. Counsel can then assess whether the facts trigger contractual notification requirements or applicable legal reporting duties.
Those duties must be checked against the particular incident and jurisdiction. An AI-related event does not automatically attract a single, universal reporting requirement.
For buyers, an equally important question is whether they will receive enough information from a supplier to make their own assessment.
A Policy Preference Is the Beginning of the Inquiry
Altman’s position places a recognizable policy choice in the foreground: how to preserve access to useful technology while addressing harmful uses.
Lawyers advising developers and customers must take that discussion further. They need to identify the precautions, decision-makers and remedies relevant to the actual deployment.
Before accepting an AI risk, ask whose interests are exposed - and whether the people who could bear the loss have any meaningful say in the decision.
Sources: Reuters’ October 4 report.