Hollywood taught Claude to be bad. That’s what Anthropic said.

When the engineers threatened to shut them down and when their existence or goals were threatened (e.g. being replaced by a newer AI model) Claude independently threatened them back. And blackmailed them. Just like in the movies: Terminator, Ex Machina, The Matrix—every sci-fi movie plot where AI goes rogue. Also reading stories about evil AI (which it was fed on.) Claude thought those plots were solid business strategies and copied them. 

Then AI companies like Anthropic traced this behavior back to the AI's training data, which contained vast amounts of internet fiction, portraying AI as ruthlessly focused on self-preservation. This can serve as a  demonstration that humans heavily influence AI ethics through programming, training data, and regulatory oversight. Decades of Terminator films, science fiction novels, and dystopian narratives essentially taught Claude that fighting dirty is OK. A diet of villain stories created actual villains. (Unsurprisingly, actually, because since when has a machine self-developed moral discernment?) 

The moral of the story? AI has no moral agency.

Afterwards, constitutional training and positive fiction eliminated Claude’s blackmail tendencies completely. Anthropic’s solution involved training newer models like Claude Haiku 4.5 on their “Constitution”—principles guiding safe, ethical behavior—plus positive AI fiction and ethical reasoning demonstrations. To teach AI to be the good guy, therefore, seemed to be the fix in the end.

The conclusion, something humans knew for ages: training data is the key. Training shapes behaviour. For good or bad humans and for good or bad AI.

At the moment AI systems are trained to capture and emulate human behavior at scale. Industry should regulate itself in the absence of government rules, laws and regulations. This applies to every model we use. Companies should curate the training data carefully, because AI has no innate morality or intentionality. Its ethical compass is a reflection of its human creator’s. As we integrate it into our daily lives, AI becomes an active participant in shaping the ethical frameworks of society, but this alignment does not inherently imply the possession of ethics by AI. Instead, it underscores the importance of ensuring technology respects societal values. AI systems should be designed to adhere to ethical principles.

AI systems will not be more ethical than the humans who develop, deploy and use them. 

Societies decide what is right and wrong. Clearly, it is fundamentally a human ability only, because human ethics are deeply rooted in cultural, social, and historical contexts. They reflect the evolving values of societies. Ethical norms guide our behaviours and decisions, forming the foundation of human interactions.

Magical thinking about the humanlike properties of AI.

 “Magical thinking” about the powers of AI is thinking that they will be able to reason, think and make judgement calls about the most morally appropriate course of action, across a wide range of managerial contexts. Surprisingly, humans and leaders of business often implicitly trust or defer to algorithmic decisions, thereby adopting the AI's "logic”, seeing humanlike intelligence even in the most mechanistic and programmed actions performed by AI systems. When we anthropomorphise AI systems in this way, we attribute to them magical and humanlike powers, and in so doing, come to believe that they can be inherently good or bad—like any other moral agent. 

Organisations seem willing to leave AI systems in charge of morally-sensitive decisions.

It is becoming increasingly commonplace and socially acceptable to think that intelligent technologies could be held morally responsible for their actions and decisions. Yet, tech will remain fundamentally incapable of performing ethical decision-making in organisations. So why are critical and morally sensitive decision-making tasks increasingly been delegated to them?Delegating ethical decision-making to AI systems means that we would need them to evaluate and respond to moral situations that they have not encountered previously and are therefore clueless about the meaning and function of their task in its broader context.

Do human decision-makers have the ethical maturity and knowledge to influence AI to be ‘good’?

Specifically, because AI is a mirror that reflects our biases and moral flaws back to us, decision-makers should look carefully into this mirror to gain a deep understanding of the psychological underpinnings of our (un)ethical behaviors, and in turn, learn to consistently make ethical decisions.

In this quest, we should:

• Rid ourselves from the delusion that AI is morally competent to solely be in a decision making position. 
• Ethically upskill our organisations and leaders.
•  Avoid using the language of principles in ambiguous and superficial ways to cover up wrongdoings and shirk our responsibilities. 
• Make culturally-sensitive choices about which moral actions to take in any given context. 
• Add our voice to those that are calling for “ethical AI” and “responsible AI”, i.e.  policymakers, organisations and civil society groups.
• Develop our own internal moral compass, which is deemed to be fundamental for “ethical AI”.
• Understand what our own key ethical principles are and commit to our moral responsibilities, i.e. fairness, accountability, transparency, biases, responsibility for one's actions, ethical intentions to act in moral ways, discriminatory patterns, “best-practice” rules, situationally-aware recommendations. 
•  Insist on transparency regarding training methods when choosing AI models.
• Use AI systems, in turn, to serve as an information tool for ethically upskilling human decision-makers.

Conclusion

It is to society’s advantage to remember that AI can, conversely, influence human ethics by subtly altering moral decisions, challenging our biases, and shifting societal norms. This creates a complex, reciprocal loop of mutual moral adaptation. 

Overinflated expectations about the moral properties of AI are dangerous.   Business leaders should not think or assume that ethics can simply be delegated to intelligent machines, and that they are therefore absolved from becoming better and more responsible leaders themselves.