Get the weekly research letter
Companies change their terms quietly. We read every version and catch what actually changed. One email a week on the changes that matter and what they mean. No account.
The policy prohibits using generative AI to generate or distribute content that facilitates spam, phishing, malware, infrastructure disruption, or circumvention of Google's abuse protections and safety filters, including prompt manipulation designed to bypass policy restrictions.
This analysis describes what Google Gemini's agreement states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability and practical outcomes may vary by jurisdiction, enforcement context, and individual circumstances. Read our methodology
This provision directly addresses adversarial use of generative AI for security attacks and model manipulation, including jailbreaking attempts, and places user-side responsibility for not circumventing Google's safety infrastructure.
Under this clause, users are prohibited from using Google's generative AI to produce content that could facilitate cyberattacks, phishing campaigns, malware distribution, or attempts to manipulate the model into violating its own policies. The agreement explicitly includes prompt-based circumvention attempts within the scope of this prohibition.
Cross-platform context
See how other platforms handle Security and Infrastructure Abuse Prohibition and similar clauses.
Compare across platforms →Monitoring
Google Gemini has changed this document before.
Receive same-day alerts, structured change summaries, and monitoring for up to 25 platforms.
"Do not compromise the security of others' or Google's services. This includes generating or distributing content that facilitates: Spam, phishing, or malware. Abuse of, harm to, interference with, or disruption to Google's or others' infrastructure or services. Circumvention of abuse protections or safety filters -- for example, manipulating the model to contravene our policies.Excerpt from Google Gemini's Google Generative AI Prohibited Use Policy
1) REGULATORY LANDSCAPE: This provision engages the Computer Fraud and Abuse Act (CFAA), which prohibits unauthorized access to and disruption of computer systems, and potentially the Electronic Communications Privacy Act (ECPA). The FTC's authority over unfair or deceptive practices including phishing and malware distribution is also relevant. State-level computer fraud and cybercrime statutes may impose additional obligations. 2) GOVERNANCE EXPOSURE: Medium for general users; High for enterprise users deploying generative AI in security-adjacent workflows, red team exercises, or penetration testing contexts. The prohibition on circumventing safety filters may affect security researchers who use prompt manipulation as part of legitimate adversarial testing, though the policy's exception clause may apply in such contexts. 3) JURISDICTION FLAGS: The CFAA applies federally across the United States. EU users face the EU Network and Information Security Directive (NIS2) in addition to national cybercrime statutes. The extraterritorial scope of the prohibition on infrastructure disruption means this provision applies globally. 4) CONTRACT AND VENDOR IMPLICATIONS: Enterprise security teams conducting authorized red team or penetration testing exercises using Google generative AI should assess whether their activities could be characterized as circumventing safety filters under this provision. Vendor agreements should clarify permissible security research use cases. 5) COMPLIANCE CONSIDERATIONS: Organizations deploying Google generative AI in security operations or research contexts should document their use cases and assess alignment with the policy's safety filter circumvention prohibition. The policy's exception clause for activities where public benefits outweigh harms may be relevant for legitimate security research, but no procedural pathway for exceptions is specified.
This provision directly addresses adversarial use of generative AI for security attacks and model manipulation, including jailbreaking attempts, and places user-side responsibility for not circumventing Google's safety infrastructure.
Under this clause, users are prohibited from using Google's generative AI to produce content that could facilitate cyberattacks, phishing campaigns, malware distribution, or attempts to manipulate the model into violating its own policies. The agreement explicitly includes prompt-based circumvention attempts within the scope of this prohibition.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by Google Gemini.