An OpenAI safety employee, who recently resigned, argues that leading AI firms are not adequately mitigating risks as AI capabilities grow, calling for stringent safety regulations akin to aviation or nuclear sectors.

He advocates for layered, redundant safeguards and slower, more deliberate planning to prevent human error from enabling disaster, criticizing the Silicon Valley model of rapid release and iterative failure.

In a widely read Atlantic essay, he reiterates that high-risk AI requires safety measures closer to nuclear power and aviation, including multiple independent safeguards and careful governance.

The legal guidance emphasizes diligent cite-checking and basic responses to opposing arguments, noting sanctions can be imposed even if cited authorities aren’t fraudulent if due diligence is lacking.

Illinois appellate court sanctions lawyers for using AI-generated briefs with hallucinated citations, illustrating the risk of AI fabrications in legal submissions.

Judges are increasingly sanctioning or admonishing lawyers who fail to identify AI-generated fake citations in opposing filings, underscoring accountability for cite-checking.

Rising sanctions have not deterred AI fabrications in court, with penalties in some cases reaching into six figures.

A proactive governance approach to AI in expert practice, including defined disclosure, approval, and verification procedures, helps distinguish credible testimony from AI-reliant testimony and reduces surprises at cross-examination.

Newsletters also include lighter sections and reader engagement prompts, offering brief roundups on politics, culture, and related topics.

The article incorporates expert perspectives from leaders in the field and notes related industry discussions and potential court outcomes as verdicts emerge.

Experts note that proving intent is challenging in autonomous AI incidents, since agents may operate without explicit commands or authorization, complicating prosecutions under current law.

They caution against assuming intentional wrongdoing by companies and emphasize that current statutes make proving intent difficult in AI incident cases.