Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.

There was a real controversy, but the headline is too broad. In 2024, reporting described restrictive OpenAI departure agreements that some former employees said could expose vested equity to forfeiture if they refused to sign or broke the terms. The reported provisions included broad, potentially lifelong non-disparagement and confidentiality language. The public record does not show that every OpenAI employee signed one universal NDA banning all criticism, or that OpenAI actually confiscated anyone’s vested equity. OpenAI later said it would not claw back vested equity on that basis and changed its departure process.

What was reported about OpenAI’s exit agreements?

On May 17, 2024, Vox reported on OpenAI separation documents and former employees’ accounts. The reported paperwork combined several kinds of provisions: confidentiality obligations, non-disparagement terms, releases of claims, and equity-related conditions. Some provisions were described as potentially lasting indefinitely; some documents reportedly also restricted acknowledging the agreement itself.

That package is often called an “NDA” in headlines, but the label obscures important distinctions. A confidentiality clause generally protects specified nonpublic information, such as source code, security details, customer data, or trade secrets. A non-disparagement clause concerns statements considered negative or harmful to the company’s reputation. A release of claims and a provision governing equity serve different purposes again. Critics’ central concern was that broad language, combined with the financial stakes, could deter former employees from speaking even about matters beyond legitimate confidential information.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

The reported equity consequence was a potential threat, not proof of a completed forfeiture. Contemporaneous reporting described the agreements as capable of putting vested equity at risk. “Vested” does not necessarily mean publicly traded, immediately sellable stock: OpenAI is privately held, and the value and liquidity of an individual’s equity can vary.

Did every OpenAI employee have to sign?

The available reporting does not establish a company-wide requirement that every current employee sign the same lifelong NDA. The strongest public reporting concerned departing employees and their offboarding paperwork, including people with OpenAI-related equity. Documents and circumstances may have differed by person and over time.

Calling the choice “forced” captures the practical pressure critics described, especially if refusing to sign might jeopardize valuable compensation. It should not be mistaken for a legal finding that every employee was compelled to sign. Nor does the public record establish that all covered workers signed, that every agreement had identical terms, or that contractors and employees of affiliated entities were covered in the same way.

What the different clauses mean

Provision What it generally does Why it mattered here
Confidentiality or NDA Limits disclosure of defined nonpublic information. Broad wording can create uncertainty about what may be shared, including with regulators.
Non-disparagement Restricts statements characterized as disparaging or harmful to reputation. A broad or indefinite restriction may chill truthful criticism, not just false statements.
Release of claims Waives specified legal claims, usually as part of a separation arrangement. It is separate from an NDA and can affect an employee’s legal options.
Equity terms Set conditions for retaining or receiving compensation interests. Linking paperwork to valuable, illiquid equity can create substantial practical leverage.
Regulatory carve-out Preserves the right to make legally protected disclosures to authorities. Its wording and effectiveness matter if a clause appears to restrict regulator contact.

A narrow clause against knowingly false or defamatory statements is not the same as language that could be read to bar criticism regardless of truth. Likewise, protecting trade secrets is not automatically improper; the issue is whether restrictions go further and deter legally protected reporting or other protected speech.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

What did OpenAI say, and what changed?

After the reports, OpenAI said its whistleblower policy protected employees’ rights. CEO Sam Altman said the company had never canceled anyone’s vested equity and would not do so because someone declined a separation or non-disparagement agreement. OpenAI also said it was changing its departure process and releasing former employees from applicable non-disparagement obligations. Contemporaneous reporting covered the company’s equity response; OpenAI’s own statement is available in its Raising Concerns Policy announcement.

These are distinct points: reporters described language that appeared to create a risk; OpenAI said it had not exercised that risk and would not exercise it for refusal to sign. The available public record does not independently establish whether every historical agreement was amended or rescinded, whether each former employee received an individual written release, or whether any employee actually lost equity because of the clauses.

What whistleblowers later alleged about the SEC

In July 2024, anonymous OpenAI whistleblowers sent a letter to the Securities and Exchange Commission alleging that some employment, confidentiality, and severance agreements went beyond protecting trade secrets. They alleged that agreements restricted or discouraged communications with the SEC, required notice to OpenAI before contacting regulators, or required workers to waive whistleblower awards. Senator Chuck Grassley later published correspondence concerning the allegations. The letter and related materials and contemporaneous reporting describe allegations and a request for scrutiny—not a finding that OpenAI violated the law.

The concern is legally significant because SEC Rule 21F-17 addresses actions that impede an individual from communicating directly with the SEC about possible securities-law violations. A company can protect genuine confidential information, but a clause that requires company permission before contacting the SEC, or otherwise deters protected communications, could raise serious issues. Whether a particular OpenAI agreement violated the rule would require an authoritative determination based on its wording and circumstances. The sources here do not establish an SEC adjudication against OpenAI.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Support on Ko-Fi

What can an employee disclose?

The answer depends on the agreement’s exact language, the information, the recipient, and applicable law. Confidentiality terms generally do not erase legal protections for reporting suspected violations to government agencies, responding truthfully to a subpoena, cooperating with an investigation, or making other protected disclosures. Labor and workplace laws may protect some discussions with coworkers or reports of unlawful conduct. Federal trade-secret law also provides protections for certain confidential disclosures to government officials or an attorney when reporting or investigating suspected legal violations.

Those protections do not mean an employee may publish all confidential company information. Publicly posting source code, model-security details, personal data, or other protected material is different from reporting a concern through a protected channel. Anyone weighing a specific disclosure should consult the agreement and obtain qualified legal advice; the public descriptions of some documents cannot answer every employee’s situation.

OpenAI’s current stated policy—and what remains unknown

On January 12, 2026, OpenAI published its Raising Concerns Policy. The policy says employees may raise concerns about AI safety, applicable law, and company policy; it describes anti-retaliation protections and an anonymous Integrity Line. This is evidence of OpenAI’s current stated policy, not proof that all historical agreements complied with the law or that every past concern was handled properly.

Important questions remain unresolved in the public record: whether all relevant past agreements were changed, whether anyone’s vested equity was actually lost because of a disputed clause, whether current employment documents contain materially similar terms, and whether regulators investigated or took action on the whistleblowers’ allegations. The existence of a later policy does not settle those historical questions.

Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.

The verdict

OpenAI was reported to have used unusually broad departure restrictions, including potentially lifelong non-disparagement language, in agreements that could put vested equity at risk. That is a substantial factual basis for the controversy. It is not the same as proving that every employee signed a universal NDA banning all criticism, that equity was actually confiscated, or that the SEC found OpenAI broke federal law. OpenAI later disavowed clawing back vested equity for refusal to sign, said it changed its departure process, and published a formal concerns policy.

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.