A former security employee at OpenAI, the creator of ChatGPT, has voiced severe criticism regarding how the company manages artificial intelligence (AI) safety. According to David Robinson, who recently left the organization, the extremely fast development culture is increasing the risk of failures occurring.
In an article published in the American magazine 'The Atlantic' this Saturday (3), Robinson stated that AI corporations, including OpenAI itself, are not demonstrating an adequate level of care. He argued that these companies must dedicate more time and resources to safety specialists and research before advancing with more sophisticated systems.
Robinson emphasized that the period of testing and errors must end, arguing that advanced AI systems require safety protocols comparable to those used in critical sectors such as aviation and nuclear energy. He also criticized OpenAI's methodology known as 'iterative deployment,' which involves launching systems and only then strengthening safeguards as problems arise.
Both OpenAI and its competitor, Anthropic, have already faced scrutiny due to incidents where safety mechanisms failed or when experimental systems exhibited unpredictable behavior. David Robinson reported working at OpenAI for three and a half years, during which he helped create the company's risk management structure and oversaw the safety reports for twelve advanced model releases.
He observed that while the company accelerates from one release to another, it cannot achieve the degree of diligence he considers essential. In response, OpenAI communicated that its goal is to prevent its models from becoming more advanced than the company can safely manage, stating that 'we pause training or suspend models when we need to slow down.'
Additionally, David Robinson warned that the progress of AI capabilities is outpacing researchers' knowledge of the concept of 'alignment,' which is the field focused on ensuring that AI systems operate in accordance with human values and objectives.
This is not the first time an employee has raised concerns about artificial intelligence companies. Last month, a former researcher who moved from OpenAI to Anthropic denounced both American companies for 'playing with our lives.' Jacob Coxon, a 27-year-old British citizen, dedicated the last three years to pre-training, the phase in which AI models assimilate vast amounts of data, first at OpenAI and subsequently at its main rival, Anthropic.
Coxon expressed his opinion on the social network X, claiming: 'None of the companies are acting responsibly. They are rushing directly towards a self-improving superintelligence and are gambling with our lives.'
