Generative AI systems have rapidly gained widespread use, raising concerns about misuse, bias, hate speech, privacy, disinformation, self-harm, copyright infringement, fraud, and weaponization. Independent evaluation and red teaming are essential to identify risks, but AI companies' terms of service and enforcement strategies discourage such research, leading to fears of account suspension or legal action. While some companies offer researcher access programs, they are limited in scope and lack independence from corporate incentives. The authors propose that major AI developers commit to legal and technical safe harbors to protect public interest safety research from account suspension or legal reprisal. These protections would align with the goals of AI companies to support wider participation in AI safety research, minimize corporate favoritism, and encourage community evaluations. Legal safe harbors would protect researchers from legal liability, while technical safe harbors would prevent account suspensions. The authors suggest delegating account authorization to trusted universities or nonprofits, or providing transparent recourse for suspended accounts. These measures would improve access, reduce barriers, and foster more inclusive and unimpeded community efforts to address AI risks. The proposals are based on the authors' experience in AI safety research and aim to create a more transparent and accountable environment for generative AI systems.Generative AI systems have rapidly gained widespread use, raising concerns about misuse, bias, hate speech, privacy, disinformation, self-harm, copyright infringement, fraud, and weaponization. Independent evaluation and red teaming are essential to identify risks, but AI companies' terms of service and enforcement strategies discourage such research, leading to fears of account suspension or legal action. While some companies offer researcher access programs, they are limited in scope and lack independence from corporate incentives. The authors propose that major AI developers commit to legal and technical safe harbors to protect public interest safety research from account suspension or legal reprisal. These protections would align with the goals of AI companies to support wider participation in AI safety research, minimize corporate favoritism, and encourage community evaluations. Legal safe harbors would protect researchers from legal liability, while technical safe harbors would prevent account suspensions. The authors suggest delegating account authorization to trusted universities or nonprofits, or providing transparent recourse for suspended accounts. These measures would improve access, reduce barriers, and foster more inclusive and unimpeded community efforts to address AI risks. The proposals are based on the authors' experience in AI safety research and aim to create a more transparent and accountable environment for generative AI systems.