Last month, OpenAI CEO Sam Altman told his followers on X that some of the guardrails would be coming off his company’s flagship product for adult users. Altman wants to allow ChatGPT to act more like a human friend, form emotional relationships with its users, and even generate erotic content—all features banned in the past because of their potential to create unhealthy attachments between users and chatbots.1
The move makes clear that many AI executives see rolling back chatbot safety features as a potential boon for their businesses. Elon Musk’s xAI, for instance, has already launched provocative or unfiltered chat features, which it appears to be using as a product differentiator.2 Users who want services unrestrained by conventional norms might flock to these platforms in the short term.
In the long term, however, corporate irresponsibility could be a major drag. Companies that produce unsafe products open themselves to vast liabilities and appeal less to the talent and customers they depend on most. Worse, they may lose consumers in the long run, as people turn against technologies that do not fit their values or that make them feel bad in other ways.
News reports regularly detail accounts of people whose mental and physical well-being are degraded by their use of AI chatbots. One corporate recruiter in Toronto, for instance, became convinced that he had superpowers after a long conversation that began when he asked ChatGPT to help him with his son’s math homework.3 To reduce risk to AI companies and increase appeal to users, avoiding pitfalls like these should be a priority for everyone in the sector.
AI Hazards
The risks taken by developers who neglect safety precautions go well beyond the consumer level. Broadly, they can be classified into issues of model use, model training, and externalities. Responsible AI products seek to avoid all three of these traps.
Scenarios like those discussed in the first section concern problems that arise from how we use a model. Developers must consider: Can users employ AI systems to generate material that is harmful to themselves or others? What moral status should the model itself have as it grows more capable?
Other ethical quandaries emerge at the training stage: Does the model use material that belongs to others without permission or compensation? Does the training process expose human workers to scarring or exploitative content?4
Finally, some risks do not directly impact the user or provider of an AI product but affect society at large. High energy use and environmental strain fall into this category, as does the so-called “existential risk” posed by a malicious or uncontrolled intelligence.
All three of these categories are real risks for those who want to lead on AI. Usage issues make products less appealing to consumers, while training and external impacts can lead to legal risks and potentially significant liabilities.
Corporate Risk
Even externalities pose substantial problems for companies that do not prioritize ethical AI development. Those who neglect their responsibilities take risks onto their balance sheets.
In most industries, this can be partially managed through insurance. A business that wants to gamble on a risky technology can take out a policy that covers its potential losses, paying a premium proportional to the hazard. But AI development presents a systematic, correlated risk that insurers are not prepared to handle. As businesses and infrastructure come to increasingly rely on AI tools, problems become more likely to ripple across firms and sectors. According to the Financial Times, the entire insurance industry currently lacks the capacity to cover major model providers.5
Already, AI companies face a wide variety of liabilities. OpenAI is being sued by writers who say its models were trained on their work without permission, and by the parents of a teenager who took his own life after confiding in ChatGPT.6,7 The unpredictable nature of large language models means they can always say or do things their creators did not intend. Companies that permit their models too much leeway risk unexpected tragedy, against which they have no protection.
Appealing to Customers and Employees
Ethical AI is not just about compliance; it is a market differentiator. According to a study from Chicago Booth, skilled workers overwhelmingly prefer to work for employers with explicit ESG commitments. Becoming a B Corp, the study found, was 77% as appealing to job seekers as the option to work from home, the most desirable benefit overall.8
In AI, where the competition for skilled talent is already ferocious, this matters enormously. Signing bonuses have reportedly reached nine figures; yet, no amount of money can retain workers who feel that their employer’s products harm society. Corporate irresponsibility could therefore drive away exactly the kind of employees companies most need—those with alternatives.
On the consumer side, the appetite for “no-limits” technology appears to be waning. Since 2022, social media engagement has declined as users grow weary of platforms that feed their worst instincts.10 The same dynamic could easily play out for AI. Products that prioritize shock value or indulgence may generate a short-term rush, but consumers will ultimately gravitate toward technologies that help them achieve their goals and align with their values.
A 2024 Deloitte survey found that 29% of AI executives—the largest single share—cited risk and compliance as the biggest obstacles to implementing agentic AI systems. But risk management is not the enemy of innovation. Responsible products are not only safer; they are more usable, more reliable, and more trustworthy. They inspire confidence, both in users and in the markets that sustain them.
behavior change 101
Start your behavior change journey at the right place
What Is It We Want?
All of this raises a fundamental question: do we want gratification, or do we want goodness? AI models that chase novelty and pleasure over genuine human improvement will eventually disappoint.
Often, we don’t give ourselves enough credit. We assume we are driven only by immediate pleasure, but, as the exodus from social media shows, we do in fact know when we’ve had enough. Products that appeal only to our basest instincts may dominate the marketplace for a while, but their appeal fades quickly compared to those that expand our capabilities and enhance our lives.
The idea goes as far back as Aristotle: the fulsome happiness that comes from acting well is better than the limited pleasure we get from fulfilling our immediate wants. Think of ethics not as a cage to restrain us, but as a guide wire to keep us aimed higher. For both the companies that create AI products and the consumers who use them, long-term health and prosperity depend on choosing what is right over what is easy. Treating responsible AI as a differentiator, not a constraint, will be the mark of the companies that lead—and last—in the decade to come.
