On the last day of the second week of Musk v. Altman, Musk's team called three witnesses.
Former OpenAI safety researcher Rosie Campbell testified that when she joined in 2021, the company had two dedicated long-term safety teams: one for AI alignment (ensuring AI behavior matches human values) and another, which she was part of, called the AGI Readiness team (preparing for superhuman AI). As OpenAI gradually became more product-focused, both teams were disbanded. About half of her team chose to leave. Campbell also detailed a specific instance: Microsoft launched a version of GPT-4 in India through Bing that had not been reviewed by OpenAI's internal safety committee. She said the model itself posed little risk, but as technology grows more powerful, "we need to set strong precedents to ensure safety processes are reliably followed." However, under cross-examination, she admitted that, in her personal judgment, Musk's xAI might be even less safety-conscious than OpenAI.
Former board member Tasha McCauley's videotaped testimony provided the board's perspective on the same events: Altman lied to the board about whether GPT-4 Turbo needed safety committee review before its India launch, claiming the legal department had already determined no review was required. McCauley said Altman fostered a "culture of lying and deception" that permeated management, leading to "a crisis every few months." She added, "The primary tool we used to fulfill our oversight duties was being dismantled."
Musk's final witness was David Schizer, former dean of Columbia Law School, who testified as an expert on nonprofit law. Lawyer Steven Molo listed Altman's behaviors described by previous witnesses and asked whether they complied with nonprofit norms. The answers were almost uniformly "no." Schizer stated, "If a CEO withholds information, that's a big problem."
On Kalshi, the probability of Musk winning the case dropped from 40% yesterday to 36%. Next week, attention will shift to Altman's own testimony.