For years, AI workers have sounded internal alarms about the risks of the technology. It took someone speaking publicly to trigger a substantive response from AI leaders.

Imagine telling the world that there’s a greater than 10% chance that your boss will destroy humanity in the next decade. That’s just what Anthropic employee Evan Hubinger did on Sept. 8, shortly after his coworker Jacob Coxon publicly resigned, warning that artificial intelligence is outpacing our ability to control it.

The same week that the AI world gathered in San Francisco to attend the Dreamforce convention, Hubinger and Coxon were not alone in fearing catastrophic consequences from a breakneck AI race. For years, AI workers have sounded internal alarms to no avail. It took someone speaking publicly to trigger a substantive response from AI leaders. But few workers are willing or able to take that bold step. Whistleblowers can face retaliation, termination, blacklisting or even a lawsuit.

Many workers are forced to sign nondisclosure agreements when they accept a job. NDAs require workers to keep silent about information employers designate as confidential. Workers who violate them can be sued for breach of contract. An estimated 88% of companies use NDAs with at least some employees and more than half of U.S. employees may be bound by one. Contractual restrictions on worker speech and mobility are especially prevalent in technology.

Article continues below this ad

Companies have legitimate reasons to protect trade secrets and proprietary assets. But NDAs often sweep far more broadly. One recent study found that many purport to stop........

© San Francisco Chronicle