AI CEOs urge lawmakers to pass AI Safety Act for risk assessments

AI leaders Dario Amodei, Sam Altman, Elon Musk and Demis Hassabis have united in warning that the latest generation of large language models could pose a serious threat to society. In a joint statement released on Tuesday, the four CEOs said the speed at which models can generate realistic text, images and audio now outpaces existing safety safeguards. They called the technology’s rapid deployment “unprecedented risk.”

The warning comes after several high‑profile incidents. In early 2024, a leaked internal memo from Anthropic revealed that its newest model could produce convincing political propaganda. At the same time, researchers at OpenAI discovered that a new version of GPT could generate “deepfake” audio that fooled human judges with 80 % accuracy. These events have sparked debate about whether current regulatory frameworks can keep up with AI innovation.

Whistleblowing agents, including former employees and independent researchers, have been publishing vulnerability reports that show how easily language models can be coaxed into generating disallowed content. Meanwhile, the term “de‑aged livers” has entered the public discourse to describe synthetic media that can alter a person’s appearance, removing signs of age or adding them back. The ability to age or de‑age faces in videos raises concerns about identity theft and political manipulation. The CEOs urged that any future model release must include transparent audits, robust testing, and a public “red‑flag” system for misuse.

Next steps include a push for federal oversight. The leaders are urging lawmakers to enact a “AI Safety Act” that would require companies to submit risk assessments and independent third‑party reviews before deploying new models. They also announced a joint partnership to create an open‑source safety toolkit that would be available to all developers. If enacted, these measures could set a global standard for responsible AI development and curb the unchecked spread of harmful content.