OpenAI’s 2024 Response to Elon Musk’s Lawsuit: Funding, Control and Openness

On February 29, 2024, Elon Musk filed a complaint against OpenAI, Sam Altman, Greg Brockman and related entities. He alleged that OpenAI departed from an agreement to develop AI for public benefit and keep its technology open, instead serving private commercial interests. These were allegations in a lawsuit, not findings by a court.

I review OpenAI’s March 5 response, signed by Greg Brockman, Ilya Sutskever, John Schulman, Sam Altman and Wojciech Zaremba. OpenAI disputes Musk’s account and says it intends to move to dismiss his claims. That statement of intent should not be confused with a court ruling or with a motion already granted.

OpenAI frames its mission as ensuring that artificial general intelligence benefits humanity. I notice that the response emphasizes distributing benefits rather than necessarily publishing the technology itself. That distinction becomes central to the disagreement about what “open” was supposed to mean.

OpenAI says its early team concluded that developing advanced AI would require far more funding and computing power than initially expected. It reports that Musk contributed less than $45 million to the nonprofit and other donors more than $90 million. These are the company’s figures in its public response.

The company publishes a 2015 email in which Musk proposes announcing a $1 billion funding commitment rather than $100 million, saying he would cover what others did not provide. A proposed commitment is distinct from the amount ultimately contributed. I read the correspondence as suggesting Musk strongly emphasized fundraising, while acknowledging that interpretation is speculative.

OpenAI says that by early 2017 the founders believed the resources required could exceed what a nonprofit could raise. I discuss compute, personnel costs and the later Microsoft relationship. My comment that Microsoft received 49% of the company is not established by the evidence shown and should not be treated as a verified ownership figure.

OpenAI asserts that it and Musk came to see a for-profit entity as necessary. It says Musk sought majority equity, initial board control and the CEO role; the other founders say they would not agree to one person having absolute control. These are OpenAI’s account of negotiations, not adjudicated findings.

The company also alleges that Musk withheld funding during the discussions and that Reid Hoffman covered a funding gap. I interpret the alleged withholding as a pressure tactic, but Musk’s motives are not established by the published excerpts. OpenAI says discussions later turned toward a possible merger with Tesla.

A forwarded February 2018 email suggested that OpenAI attach to Tesla as its “cash cow,” and Musk endorsed the proposed route as a way to compete with Google. In context, the writer describes Tesla as a potential source of funding for OpenAI. My suggestion that OpenAI would itself be Tesla’s cash cow reverses that intended relationship.

I consider whether the proposed Tesla arrangement supports OpenAI’s position that Musk contemplated a for-profit path. It is relevant to the dispute, but it does not by itself settle what the parties agreed or establish that Musk accepted every later decision about proprietary technology.

OpenAI says Musk left in February 2018 and planned a competitor within Tesla. It also cites a December 2018 message in which he argued that the venture needed billions of dollars per year. I distinguish his expressed view about financing from a definitive agreement that all future research should be closed.

OpenAI argues that broadly available tools can deliver the benefits of AI even if the underlying science is not always published. I question how benefits can be distributed fairly when a single organization controls access and conditions. That is a governance concern, separate from whether a particular contract was breached.

To support its interpretation of openness, OpenAI displays a 2016 exchange with Ilya Sutskever. Sutskever wrote that as the organization got closer to building AI it might make sense to share less science, while seeking to spread the resulting benefits. The published reply from Musk was a brief “yup.”

I question how much can be inferred from that one-word reply. It is evidence of the exchange, but the published excerpts alone do not establish exactly what Musk read, all the surrounding discussions or the legal meaning of the response. My suggestion that he may have skimmed the message is speculation.

OpenAI says Musk eventually started a competitor and sued after the organization made progress without him. I speculate that Musk may have personal or competitive motives, while also taking seriously his objections to OpenAI’s change in structure. Neither the company’s characterization nor my inference proves his motives.

I return to the original 2015 email to Greg Brockman, with Altman copied. Musk recommends a public-facing announcement and a separate more detailed recruiting version, then argues for a larger funding commitment. He also discusses compensation involving possible equity in other entities; I say I do not know precisely how that proposal would have worked.

Another displayed message, dated February 1, 2018, forwards an email whose author is redacted. Musk agrees with its argument that linking OpenAI to Tesla offered a possible way to compete with Google. I consider the writer’s identity, but the redacted material does not establish who that person was.

The anonymous writer argues that frontier AI research is expensive and that compute, data and infrastructure may be crucial. The writer suggests published advances could be adopted by a better-resourced competitor. These are forecasts and strategic arguments in the email, not proof that compute alone is sufficient for AGI.

The email weighs a for-profit pivot as a possible way to generate revenue and investment, but warns that creating a product could distract from research and give investors influence. It then proposes a connection to Tesla as a way to secure resources. I discuss the argument without treating its predictions as established outcomes.

I observe that Google’s Transformer research was openly published and later became influential across the industry, including at OpenAI. That history complicates a simple story that only one organization used the research; my broader remarks are commentary on the strategic value of published work.

A separate December 2018 email has Musk assessing OpenAI’s chances of being relevant against Google and DeepMind at zero without a dramatic change in execution and resources. The condition matters: it was not an unconditional prediction that the organization could never succeed under any circumstances.

I read another redacted email warning that open-sourcing powerful AI is not necessarily a solution to safety and may create risks. Its author references an essay by Scott Alexander. The identity of the redacted sender is not established by the excerpt, and the risk scenario is an argument rather than an observed outcome.

Sutskever’s January 2016 email addresses a hypothetical rapid takeoff in which unsafe AI might be easier to build than safe AI. He suggests that openness might need to decrease closer to advanced AI, while sharing research could remain useful earlier. Musk’s brief reply appears beneath it in the material OpenAI published.

I play devil’s advocate about whether Musk’s “yup” should be read as agreement with every sentence. I say I am not a lawyer and do not know its legal significance. The exchange provides context for a disputed history, but does not on its own resolve the complaint or the parties’ wider disagreement.

The video closes by asking whether OpenAI can obtain dismissal and how viewers interpret the emails. As of this March 2024 discussion, the complaint and public response presented opposing positions; the video does not establish a legal outcome.