Elon Musk’s lawsuit puts OpenAI’s security record under the microscope

Elon Musk’s legal efforts to end OpenAI may depend on how his for-profit company promotes or hinders the startup’s efforts to ensure that humans benefit from artificial intelligence.

On Thursday, a federal court in Oakland heard from a former employee and board member that the company’s efforts to push AI products into the market undermined its commitment to AI safety.

Rosie Campbell joined the company’s AGI planning team in 2021, and left OpenAI in 2024 after his group was dissolved. Another security team, the Super Alignment team, was shut down immediately.

“When I joined it was very research and it was very common for people to talk about AGI and security,” he said. “Over time it became more of a product-focused organization.”

When questioned, Campbell admitted that a lot of money is needed for the lab to develop AGI, but he said that creating super-intelligent computers without proper security measures would not fit the mission of the organization he joined in the first place.

Campbell pointed to what happened when Microsoft sent the company’s version of GPT-4 to India via the Bing search engine before it was reviewed by the company’s Deployment Safety Board (DSB). The model didn’t present a major risk, he said, but the company needs to “establish strong models as the technology matures. We want to have good security measures that we know are reliably followed.”

OpenAI advocates Campbell also admitted that “in theory,” OpenAI’s security strategy is superior to that of xAI, the AI ​​company Musk founded that was acquired by SpaceX earlier this year.

OpenAI releases an overview of its models and parameters security framework publicly, but the company declined to comment on its current AGI performance. Dylan Scandinaro, their head of Preparedness, was hired at Anthropic in February. Altman he said the hired hand would let him “sleep well tonight.”

The deployment of GPT-4 in India, however, was one of the red flags that led the non-profit organization OpenAI to briefly fire CEO Sam Altman in 2023. This happened after employees including the then chief scientist Ilya Sutskever and CTO Mira Murati complained about Altman’s management style. Tasha McCauley, a board member at the time, testified to concerns that Altman was not forthcoming enough with the group to make his unusual appearance work.

McCauley also discussed a multi-reports example Mr. Altman misled the group. In particular, Altman lied to one board member about McCauley’s intention to fire Helen Toner, a third board member who published a white paper that included criticism of OpenAI’s security policy. Altman also failed to inform the group of the decision to establish ChatGPT publicly, and members were concerned about not disclosing potential conflicts.

“We are a not-for-profit organization and our job was to manage the profits under our control,” McCauley told the court. “Our main way of doing this was doubt.

However, the idea to boot Altman came immediately as a charitable donation to the company’s employees. McCauley said that after OpenAI staff began siding with Altman and Microsoft worked to reform the existing organization, the organization changed, with members who opposed Altman resigning.

The apparent failure of a non-profit organization to attract a profit-making organization goes directly to Musk’s case that the transformation of OpenAI from a research organization into one of the largest companies in the world has violated the guarantee agreement of the founders of the organization.

David Schizer, the former Dean of Columbia Law School who is being paid by Musk’s team to be a key witness, echoed McCauley’s concerns.

“OpenAI has emphasized that a major part of its mission is security and will prioritize security over profit,” Schizer said. “Part of this is looking at safety regulations more seriously, if something needs to be reviewed, it needs to be done.”

With AI already entrenched in the for-profit industry, the issue extends beyond one lab. McCauley said that the lack of internal control at OpenAI must be due to strong government regulation of advanced AI – “(if) it all comes down to one CEO making these decisions, and we’re at risk for the public, then it’s too little.”