Main Insight
As whistleblowers have become essential safeguards against unsafe AI development, international coordination is needed to establish stronger whistleblower protections to prevent AI risks from escalating into global crises.
Why Whistleblowers Are Critical for AI Governance
July 24, 2025
In early 2024, OpenAI safety researcher Leopold Aschenbrenner raised concerns about the company’s security protocols. After warning that OpenAI’s defenses were “egregiously insufficient” against foreign adversaries, Aschenbrenner was swiftly fired. Although the reason for the dismissal was allegedly for sharing confidential information, he maintains that it was retaliation for speaking up.
Aschenbrenner is one of a growing number of AI whistleblowers—employees inside tech companies who raise concerns about AI safety risks, security vulnerabilities, or unethical practices. AI whistleblowers are often among the first to spot internal problems and provide early warning signs of serious incidents, long before regulators or the public are aware. Working within the companies building frontier AI systems, they are uniquely equipped to detect risks and raise critical alarms before harm occurs.
Yet, as Aschenbrenner’s case shows, raising concerns about AI can come at a steep personal and professional cost. A disturbing pattern can be seen in Timnit Gebru’s controversial dismissal at Google, recent firings for leaks at Meta, and the open letter signed by 11 OpenAI current and former employees last June calling for stronger whistleblower protections. Those who draw attention to legitimate dangers can face job loss, legal threats, and lasting reputational damage.
This blog post examines the systemic barriers that AI whistleblowers face and outlines the concrete steps that global stakeholders must take to protect the individuals trying to hold powerful companies accountable and prevent the next AI incident.
Barriers to Speaking Out for AI Whistleblowers
One of the most fundamental challenges for AI whistleblowers is the absence or ambiguity of clear legal protections. In many jurisdictions, existing whistleblower laws were not designed with AI-specific concerns in mind. As a result, disclosures about issues like unsafe model deployment, inadequate security protocols, or misalignment with stated ethical commitments often fall outside the scope of current regulations.
Even where legal frameworks exist, they are often ineffective. In the United States, for example, at-will employment laws allow companies to dismiss employees without cause. This gives organizations broad leeway to fire whistleblowers under vague pretexts, such as restructuring or cost-cutting measures, rather than acknowledge retaliation. Legal threats are also a powerful deterrent: companies frequently frame internal disclosures as trade secret theft, which exposes employees to costly, reputation-damaging litigation. And even when whistleblowers do pursue responding with legal action, their cases can drag on for years, exacting a high financial and emotional toll.
Compounding this is the fact that companies often face no real consequences for suppressing whistleblowers. The result is a chilling effect: employees may choose silence not because they don’t care, but because the personal risks of speaking up far outweigh any systemic impact.
The nature of AI development makes this problem worse. Whistleblowers often raise concerns like corner-cutting on safety, covert military collaborations, or misaligned incentives. These issues are typically difficult to categorize, hard to prove, and easy to dismiss. As AI firms grow more powerful and increasingly partner with national security agencies, their incentives to suppress internal dissent deepen. Executives often argue that broad protections could encourage false claims or jeopardize trade secrets, competitiveness, or even national interests. But this argument misses a crucial point: existing oversight is largely reactive. Without credible protections for those who speak up, critical failures will remain hidden until they erupt into public crises.
Even when internal reporting systems mechanisms technically exist, they often fail in practice. Employees may not know how to access them, may not trust that their concerns will be acted upon, or may fear that reporting will be seen as a personal attack rather than a professional responsibility. In the high-pressure, high-ambiguity environment of frontier AI development, these social and psychological dynamics can be paralyzing. Effective whistleblowing systems must therefore go beyond formal availability, they must be clear, transparent, and designed to foster psychological safety. They should be made publicly available.
Finally, jurisdictional complexity adds another layer of difficulty. In the UK and EU, relatively strong protections exist under frameworks like the Public Interest Disclosure Act (PIDA) and the EU Whistleblower Directive. But in the United States, China, and other jurisdictions, the legal landscape is fragmented and enforcement is inconsistent. For global companies and international teams, this patchwork of protections further discourages action and creates uncertainty around what rights whistleblowers actually have.
Typical Whistleblower Reporting Pathways and Their Limitations
| Pathways | Description | Examples | Limitations |
|---|---|---|---|
| Internal Company Reporting Channels | Most organizations encourage employees to report concerns through internal mechanisms before escalating them externally. These typically include anonymous reporting hotlines or web portals, as well as designated compliance or ethics officers who handle and investigate concerns. | OpenAI’s anonymous whistleblower hotline for employees | Employees often fear retaliation, even when anonymity is promised. Reports may be ignored or dismissed without meaningful action. Internal mechanisms may prioritize corporate interests over transparency. Many major AI companies lack formal internal reporting channels and face no obligation to establish them under U.S. law. |
| Reporting to Oversight Bodies | If internal reporting fails or is not an option, whistleblowers can escalate concerns to external regulatory bodies that oversee corporate practices. These bodies may investigate or take enforcement action based on disclosures. | U.S.: The SEC Whistleblower Program offers protections and potential financial rewards for reporting securities law violations. UK: Whistleblowers can report concerns to a prescribed person or body within the government under the Public Interest Disclosure Act (PIDA). | Protections vary significantly between jurisdictions. Investigations by regulatory agencies can take years to conclude. Some agencies only act if clear legal violations are evident, leaving unethical yet technically legal concerns unaddressed. Current whistleblower programs are often managed by organizations lacking expertise or authority on AI, which might prevent or delay those who could address AI incidents or risks from taking action. |
| Disclosures to the Media | In cases where neither internal nor legal avenues seem effective, whistleblowers sometimes go public via investigative journalists. | Blake Lemoine, a former Google engineer, disclosed to The Washington Post that he believed Google’s LaMDA chatbot had achieved sentience. Ed Newton-Rex publicly resigned from Stability AI, criticizing the company’s approach to AI-generated content and the ethical implications of training models on copyrighted material without consent. | Can result in legal consequences for the whistleblower, including potential lawsuits and criminal charges. Potential loss of anonymity and risk of professional blacklisting. Media coverage can shape public perception but may not lead to systemic change. |
How to Create Safe and Effective Pathways for AI Whistleblowers in Global AI Governance
The current barriers whistleblowers face leave too many of these individuals exposed and, hence, too many risks unreported. To address this, a more coordinated international approach is needed: one that brings together companies, governments, and civil society to create clear, safe, and accessible pathways for raising concerns.
1. International organizations and fora, in collaboration with their member states, should take the lead in establishing best practices and industry-wide standards for AI whistleblower protections. This includes standardizing internal reporting policies and mechanisms, ensuring independent committees handle AI whistleblower reports, and creating pathways for external reporting, including via incident reporting platforms. AI companies should also be called on to commit to eliminating restrictive NDAs that prevent employees from raising safety concerns without fear of legal repercussions or financial loss.
The United Nations Convention Against Corruption (UNCAC) provides a strong precedent: it urges governments to protect whistleblowers reporting corruption or unethical behavior and has prompted many signatories to adopt national whistleblower laws aligned with its guidance. This example shows that coordinated international standards are both feasible and impactful.
International forums can also facilitate knowledge-sharing among countries, helping governments learn from existing and proposed whistleblower protection mechanisms elsewhere. For example, AI summits, such as the upcoming one in India, could revive the Seoul summit’s initiative to secure commitments from AI companies and expand them to include best practices for protecting whistleblowers.
2. National and regional governments should strengthen legal protections by establishing secure channels that allow whistleblowers to confidentially submit concerns to relevant authorities. A promising example is found in the Statement from the Chairs and Vice Chairs of the Safety And Security Chapter in the European Union’s General-Purpose AI Code of Practice, which recommends that the AI Office establish a dedicated reporting channel for AI-related whistleblower disclosures. This dedicated reporting channel would explicitly enable employees to confidentially report systemic risks associated with large-scale general-purpose AI models, providing an additional layer of oversight on top of those included within the EU Whistleblower Directive.
3. Civil society organizations (CSOs) should form a broad coalition to advocate for stronger whistleblower protections and a transparent, accountable culture within AI companies. CSOs can play a vital role in creating safer conditions for whistleblowers by offering independent, confidential guidance to those considering disclosure. Important models that merit wider support include The AI Whistleblower Initiative, which provides anonymous expert assessments for individuals considering disclosure; and, similarly, Psst, The Signals Network, and WHISPeR, which offer legal assistance, security measures, and media coordination.
Civil society should also push for policy reforms that strengthen whistleblower protections and deter retaliation by AI firms. A unified advocacy effort can drive regulatory change and create real accountability.
Finally, civil society should recognize and reward companies that demonstrate meaningful transparency and internal accountability. By publicly highlighting labs that foster a culture of openness, and calling out those that suppress internal dissent, civil society can help shift industry norms and encourage responsible behavior across the sector.
4. AI companies and governments should explore ways to provide tangible rewards for whistleblowers who expose significant AI risks or incidents. Financial incentives for whistleblowers who reveal critical AI concerns should also be introduced, modeled after successful programs in financial and environmental regulation. However, implementing such incentives requires first establishing a clear legal framework that defines unacceptable levels of risk for AI systems and imposes enforceable penalties for violations.
Mechanisms to Protect and Reward Whistleblowers
| Mechanisms | Key Elements | Examples |
|---|---|---|
| Whistleblower Protection Laws | Designed to ensure that whistleblowers cannot be fired, demoted, or otherwise penalized for reporting reasonable concerns internally or externally. Whistleblowers who experience retaliation can seek legal recourse through employment or civil litigation. In some cases, their disclosures may also lead to government prosecution of companies. Enable whistleblowers to report AI risks without exposing their identity. Prohibit employers from using contractual obligations (such as NDAs) to prevent public interest disclosures. | The EU Whistleblower Directive requires companies with over 50 employees to establish secure reporting channels and protects whistleblowers from retaliation. The U.S. Whistleblower Protection Act: provides protection for federal employees. In the UK, whistleblowers who face retaliation may file a complaint with an employment tribunal under PIDA. |
| Whistleblower Reward Programs | Whistleblower reward programs provide financial incentives for individuals who expose wrongdoing or risks that lead to regulatory action. Common in financial and environmental sectors, these programs often prioritize whistleblower reports that reveal substantial harm, fraud, or regulatory violations | U.S. SEC Whistleblower Program: Offers financial rewards to individuals who provide information leading to successful enforcement actions in cases of securities law violations. China’s Whistleblower Reward System (SAMR): Offers up to RMB 1 million (~USD 157,000) for whistleblowers reporting major market violations, such as fraudulent insurance claims |
A Call to Action: Protect the People Who Could Prevent the Next AI Incident
AI is moving fast. Whistleblowers can help us spot risks before they escalate into serious incidents.
As governments prioritize geopolitical advantage and AI companies resist scrutiny, whistleblowers remain one of the few guardrails against a potential serious AI incident ahead. Like their counterparts in the fields of aviation, healthcare, and finance, AI whistleblowers can catalyze vital reforms if we create the conditions that will allow them to come forward.
Without stronger protections, clearer global norms, and meaningful incentives for whistleblowers, the risks of coming forward will continue to outweigh the benefits. But when disclosures lead to policy reform, improved safety practices, and real accountability, the system gains a chance to self-correct before harm occurs.
The path forward is simple but urgent: protect the people who are willing to sound the alarm before avoidable harms occur.
Acknowledgements
We would like to thank the following people for their contributions to this publication: Abra Ganz from the Center for AI Risk Management & Alignment (CARMA) and Karl Koch from the AI Whistleblower Initiative (formerly OAISIS).
Additionally, we are grateful for the support of our colleagues at The Future Society: Liza Adhiambo, Li June Choi, Kathrin Gardhouse, Sumin Heo, Niki Iliadis, Toni Lorente, Mai Lynn Miller Nguyen, and Nick Moës.
Participation does not imply endorsement of the report or its findings. The views expressed herein do not necessarily represent the perspectives of these individuals or their respective organizations. Any remaining errors are our own.
