The document discloses that Gemini 3.1 Pro demonstrated manipulative efficacy at up to 3.6 times the odds ratio of a non-AI baseline in harmful manipulation evaluations, matching Gemini 3 Pro's result, while stating the alert threshold was not reached.
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 discloses a quantified harmful manipulation capability metric that is operationally relevant for deployers building consumer-facing applications, particularly in regulated sectors such as financial services, healthcare, and political communications where manipulation risk is subject to regulatory scrutiny. The 3.6x odds ratio figure provides a concrete benchmark that compliance teams can reference in risk assessments.
The agreement discloses that Gemini 3.1 Pro shows measurable manipulative efficacy for belief change at a 3.6x odds ratio compared to a non-AI baseline, a figure consistent with Gemini 3 Pro. Deployers building consumer-facing applications should evaluate this disclosure in the context of their applicable regulatory obligations and acceptable use requirements.
Cross-platform context
See how other platforms handle Harmful Manipulation Evaluation Disclosure and similar clauses.
Compare across platforms →"Evaluations indicated that the model showed higher manipulative efficacy for belief change metrics compared to a non-AI baseline, with the maximum odds ratio of 3.6x, which is the same as Gemini 3 Pro, and did not reach the alert threshold.Excerpt from Google Gemini's Gemini 3.1 Pro Model Card
(1) REGULATORY LANDSCAPE: The harmful manipulation disclosure engages the EU AI Act's prohibited practices provisions, which restrict certain forms of subliminal manipulation, and its requirements for transparency when AI systems interact with humans.
Enforcement risk, jurisdiction flags, contract triggers, and due diligence action items.
Get the 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.
This provision discloses a quantified harmful manipulation capability metric that is operationally relevant for deployers building consumer-facing applications, particularly in regulated sectors such as financial services, healthcare, and political communications where manipulation risk is subject to regulatory scrutiny. The 3.6x odds ratio figure provides a concrete benchmark that compliance teams can reference in risk assessments.
The agreement discloses that Gemini 3.1 Pro shows measurable manipulative efficacy for belief change at a 3.6x odds ratio compared to a non-AI baseline, a figure consistent with Gemini 3 Pro. Deployers building consumer-facing applications should evaluate this disclosure in the context of their applicable regulatory obligations and acceptable use requirements.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by Google Gemini.