"AI Firms Are Gambling With Our Lives": Researchers Leaving Over Safety Concerns

Amid growing concerns over the safety of artificial intelligence (AI), employees at leading AI companies and related organizations are suffering from mental stress over whether their work poses a serious threat to society, the British daily Financial Times reported on September 22.

According to sources, several staff members at the UK AI Safety Institute (AISI), an organization that independently evaluates frontier AI models, have taken medical leave and sought counseling due to mental distress.

Sources said these employees are experiencing low morale and burnout due to tight testing schedules for upcoming AI models, as well as concerns over the rapid spread and functional advancement of new AI technologies.

Employees at leading companies such as Anthropic, OpenAI, and Google DeepMind are voicing similar concerns.

A DeepMind employee confessed that making AI safer feels like an "impossible mission."

Referring to the backlash in April when Google executives and employees protested the use of AI models for classified U.S. Department of Defense military operations, the employee said, "Working on AI systems generally carries a (mental health) risk because you don't know how the technology will be used."

"Many people are confused and are quitting or taking medical leave for mental health issues," the employee added. "There is also a feeling of losing control and a feeling that we are building something genuinely harmful."

According to industry insiders, AI researchers are deeply concerned that AI software is being applied to dangerous areas such as energy grids and war zones even though internal corporate or political safety measures are insufficient.

A former AISI researcher said that as AI rapidly develops capabilities to perform complex tasks such as finding flaws in code that humans missed or generating viruses that do not occur naturally, workers in cybersecurity and biosecurity fields are particularly shocked.

"There's a sense that there are reckless elements inherent in the way the technology is being deployed," the former researcher said. "So there are constant concerns, like 'Is there anything that can stop this?' and 'Are we watching the world burn?'"

The FT reported that more than 12 high-level researchers have resigned from the three major U.S. leading companies—OpenAI, Anthropic, and DeepMind—over the past two years, expressing concerns over safety.

Earlier this month, when Anthropic AI developer Jacob Coxon resigned, he warned that Anthropic and OpenAI are "gambling with our lives."

Mrinank Sharma, a former AI safety researcher at Anthropic, resigned in February this year to study poetry, writing in a farewell letter to colleagues, "The world is dangerous. I have repeatedly witnessed how difficult it is to act in accordance with our values."

Hieu Pham, a former OpenAI employee, also quit this year and returned to his native Vietnam, stating, "Building something extremely intelligent comes with a price. The degradation of mental health is real, miserable, scary, and dangerous."
※ Please note: This article was translated by AI and may contain errors.