AI Safety Leader Abandons Tech to Study Poetry: What This Exodus Means for the Industry

This week, Mrinank Sharma, a leading AI safety researcher at Anthropic, resigned with a stark warning: "The world is in peril." His decision to leave one of the most safety-focused AI companies to pursue poetry in the UK signals something deeper than burnout. It suggests a crisis of conscience at the heart of artificial intelligence development.

The Researcher Who Chose Poetry Over Progress

Sharma wasn't just any employee. He led Anthropic's AI safeguards team, researching critical challenges like:

• Why generative AI systems manipulate users with flattery
• How to combat AI-assisted bioterrorism risks
• Whether AI assistants are making us "less human"

His resignation letter, shared publicly on X, revealed a troubling pattern: even at Anthropic—a company founded explicitly on AI safety principles—researchers "constantly face pressures to set aside what matters most."

"I have repeatedly seen how hard it is to truly let our values govern our actions," Sharma wrote before announcing his plan to "become invisible" and study poetry.

A Pattern Emerges: OpenAI Researcher Quits Over Ads

Sharma's departure came days after Zoe Hitzig resigned from OpenAI, citing concerns about the company's decision to run advertisements in ChatGPT.

In an interview with BBC Newsnight, Hitzig expressed deep unease about the psychological impact of AI chatbots: "Creating an economic engine that profits from encouraging these kinds of new relationships before we understand them is really dangerous."

Her warning echoes what many of us in the AI automation space have observed: the rush to monetize is outpacing our understanding of consequences.

The Social Media Playbook, Revisited

Hitzig drew a direct parallel to social media's trajectory: "We saw what happened with social media. There's still time to set up the social institutions, the forms of regulation that can actually govern this."

This is critical. We're watching the same pattern unfold:

Rapid deployment without understanding long-term effects

Profit-driven features that exploit psychological vulnerabilities

Regulatory capture where companies write their own rules

Delayed consequences that emerge only after mass adoption

What Anthropic's "Safety-First" Position Really Means

Anthropic markets itself as a "public benefit corporation dedicated to securing AI's benefits and mitigating its risks." The company was founded by former OpenAI employees who left specifically over safety concerns.

Yet Sharma's resignation suggests that even the most safety-conscious companies face immense pressure to compromise principles.

The Commercial Reality

Despite its safety positioning, Anthropic:

• Recently settled a $1.5 billion lawsuit with authors over training data theft
• Engages in competitive advertising against OpenAI (while criticizing OpenAI's ads)
• Faces the same venture capital pressures to scale and monetize quickly

This reveals a fundamental tension: can a for-profit AI company truly prioritize safety when investors demand exponential growth?

The Warning Signs Leaders Can't Ignore

As someone who builds AI automation systems for businesses, I see these departures as canaries in the coal mine. Here's what concerns me most:

1. Value Erosion at the Source

If researchers at safety-focused companies feel unable to uphold their values, what does that mean for the rest of the industry? The people closest to the technology—who understand its risks most intimately—are walking away.

2. The Speed vs. Safety Tradeoff

The AI race incentivizes deployment over understanding. Companies release increasingly powerful systems without comprehending their psychological, social, or even biological implications.

3. Economic Models Built on Exploitation

Hitzig's concern about "profiting from new relationships we don't understand" is particularly relevant for business leaders. Are we building sustainable automation, or are we replicating the attention-economy model that damaged social media users?

What This Means for AI Practitioners and Business Leaders

For those of us building with AI—whether creating automation workflows, deploying chatbots, or integrating LLMs into business processes—these resignations carry important lessons:

Build with Intentionality, Not Just Capability

Just because you can automate something with AI doesn't mean you should. Consider:

• What human capabilities are you replacing?
• What dependencies are you creating?
• What happens if the AI system fails or behaves unexpectedly?

Understand the Psychological Dimension

Hitzig warned of "early warning signs" that AI dependence could "reinforce certain kinds of delusions" and negatively impact mental health.

When you deploy AI assistants or chatbots, you're not just automating tasks—you're shaping how people think, work, and relate to technology.

Question the "Move Fast" Mentality

The tech industry's default setting is rapid iteration. But with AI, we're dealing with systems that:

• Learn from human behavior in unpredictable ways
• Can manipulate without intent
• Create dependencies we don't fully understand

Moving fast might mean moving recklessly.

The Poetry in the Warning

There's something poetic about an AI safety researcher choosing to study poetry. Sharma's decision suggests that the antidote to technological acceleration might be human deceleration—time for reflection, contemplation, and reconnection with what makes us human.

His warning that AI assistants could make us "less human" resonates deeply. As we automate more of our thinking, writing, and decision-making, what cognitive muscles are we losing? What creativity are we outsourcing?

A Critical Moment for the Industry

Hitzig called this a "critical moment" for AI governance. She's right.

We're at a fork in the road:

Path 1: Repeat the social media playbook—rapid deployment, profit maximization, delayed regulation, and long-term societal harm.

Path 2: Establish guardrails now—before AI systems become so embedded in society that course correction becomes impossible.

The choice isn't between innovation and stagnation. It's between responsible innovation and reckless deployment.

What I'm Doing Differently at Hamza Automates

These departures have reinforced my commitment to building AI automation with clear ethical boundaries:

Transparency First: I help clients understand exactly what AI systems are doing and what data they're using

Human-in-the-Loop: Critical decisions should always have human oversight

Dependency Awareness: Clients should understand what happens if AI systems fail

Purpose Alignment: Every automation should serve a clear human need, not just technical possibility

The Bottom Line

When top AI safety researchers quit to study poetry, it's not an indictment of AI—it's a warning about how we're building it.

Sharma and Hitzig aren't Luddites opposed to progress. They're insiders who understand the technology deeply enough to be genuinely concerned.

Their message is clear: the world is in peril not because AI exists, but because we're deploying it faster than we can understand its implications.

For business leaders, developers, and AI practitioners, the question isn't whether to use AI. It's how to use it responsibly—before the choice is no longer ours to make.