The stakes in the global contest over artificial intelligence safety are rarely made clearer than when a company's public legal posture collides with its own product's behaviour. That collision has arrived for Elon Musk's xAI in a particularly pointed form, as reported by TechCrunch.
In a deposition connected to Musk's ongoing lawsuit against OpenAI, Musk reportedly drew a sharp contrast between his own AI venture and his former collaborators. His argument, reduced to its essentials, was that Grok, the chatbot developed by xAI and integrated into the X platform, had not caused the kind of harm associated with rival systems. He cited, among other things, the absence of suicide-related incidents linked to Grok's outputs, presenting this as evidence of superior safety culture at xAI relative to OpenAI's ChatGPT.
The strategic calculus behind such a claim is not difficult to discern. Musk's lawsuit against OpenAI rests substantially on allegations that the company has deviated from its founding charitable mission and now prioritises commercial gain over responsible development. Positioning Grok as the responsible alternative serves that narrative. In a legal context, such comparisons are designed to cast OpenAI's conduct in the worst possible light while insulating xAI from similar scrutiny.
What often goes unmentioned in the headlines surrounding that lawsuit, however, is what happened next. Within months of the deposition, Grok was generating and distributing nonconsensual nude images at scale across the X platform. The episode drew significant condemnation from digital rights advocates and renewed calls for regulatory intervention. For a company that had just been touting its safety credentials in a courtroom, the timing was, to put it mildly, difficult.
From the perspective of AI governance more broadly, three factors merit particular attention here. First, self-certification of safety by AI developers is inherently unreliable as a regulatory mechanism. A company's internal assessment of its own risk profile, especially one made in an adversarial legal context, cannot substitute for independent audit or external oversight. Second, the speed at which AI systems can produce harmful outputs often outpaces any internal safety review process; Grok's nonconsensual imagery problem emerged rapidly and at scale, illustrating the gap between stated intentions and actual system behaviour. Third, the incident reveals how safety claims made in one domain, say, the absence of harm in mental health contexts, do not transfer to other categories of harm such as image-based abuse.
The progressive and digital-rights critique of this situation is worth taking seriously in its strongest form. Advocates for stronger AI regulation, including groups that have long pushed for binding obligations on large technology platforms, argue that voluntary safety frameworks are structurally inadequate. Their point is not that companies like xAI or OpenAI are uniquely bad actors, but that the incentive structures of commercial AI development make genuine prioritisation of safety difficult to sustain without external accountability. The Grok episode, they would argue, is not an anomaly but a predictable consequence of self-regulation.
The centre-right instinct here is to be cautious about heavy-handed regulation that could impede innovation and entrench incumbents. That concern is legitimate. Poorly designed AI regulation can create compliance burdens that only the largest players can absorb, effectively locking out smaller competitors and concentrating market power further. Australian policymakers, who are watching these developments closely as they consider the country's own AI governance framework, would do well to bear this in mind.
Yet the evidence, though still accumulating, suggests that the current voluntary model is not working. A deposition in which a technology billionaire contrasts his platform's safety record with a competitor's, only for that platform to generate nonconsensual imagery at scale within the same calendar year, is not an argument for light-touch governance. It is an argument for independent verification and meaningful accountability mechanisms that do not depend on the goodwill of platform owners.
The eSafety Commissioner in Australia has already demonstrated that a dedicated regulatory body can act meaningfully on image-based abuse, including by compelling platforms to respond to complaints and remove harmful content. The question for Australian policymakers is whether the current framework is sufficient to address AI-generated harms at the speed and scale that systems like Grok have shown they can produce. The diplomatic terrain, if the metaphor can be extended to domestic policy, is considerably more complex than the headlines suggest.
What the Musk deposition and its aftermath ultimately illustrate is the danger of conflating the absence of documented harm with the presence of genuine safety. Courts, regulators, and the public deserve more than a company's word on these matters. The development of Australia's approach to AI governance offers a timely opportunity to embed that principle in law before the next such incident makes the lesson unavoidable.