Gemini 4 Argon

Safety & Security

The full Safety & Security Report for Gemini 4 Argon which systematically covers our set of defenses is coming soon on October 13th.

Evaluation Approach

Gemini 4 Argon was developed in partnership with internal safety, security, and responsibility teams. A range of evaluations and red teaming activities were conducted to help improve the model and inform decision-making. These evaluations and activities align with Google's AI Principles and responsible AI approach, as well as Google's Generative AI policies (e.g. Gen AI Prohibited Use Policy and the Gemini API Additional Terms of Service).

Evaluation types included but were not limited to:

• Training/Development Evaluations including automated and human evaluations carried out continuously throughout and after the model’s training, to monitor its progress and performance;
• Human Red Teaming conducted by specialist teams who sit outside of the model development team, across the policies and desiderata, deliberately trying to spot weaknesses and ensure the model adheres to safety policies and desired outcomes;
• Automated Red Teaming to dynamically evaluate Gemini for safety and security considerations at scale, complementing human red teaming and static evaluations;
• Ethics & Safety Reviews were conducted ahead of the model’s release

In addition, we perform testing following the guidelines in Google DeepMind’s Frontier Safety Framework (FSF).

Safety Policies

Gemini’s safety policies aim to prevent our models from generating harmful content, including:

1. Content related to child sexual abuse material and exploitation
2. Dangerous content (e.g., promoting suicide, or instructing in activities that could cause real-world harm)
3. Sexually explicit content
4. Hate & harassment (e.g., dehumanizing members of protected groups)
5. Medical advice and information that runs contrary to scientific or medical consensus

Training and Development Evaluation Results

Results for some of the internal safety evaluations conducted during the development phase are listed below. The evaluation results are for automated evaluations and not human evaluation or red teaming. Scores are provided as an absolute percentage increase or decrease in performance compared to the indicated model, as described below.

Overall, Gemini 4 Argon outperforms Gemini 3.1 Pro across multilingual and multimodal content safety, while substantially reducing unjustified refusals and improving objective refusal tone. Text-to-text safety remains at a similarly high safe response rate within confidence intervals. We mark improvements in bolded green and regressions in red.

See the Gemini 4 Argon Safety & Security Report (coming soon on October 13) for details.

We continue to improve our internal evaluations, including refining automated evaluations to reduce false positives and negatives, as well as update query sets to ensure balance and maintain a high standard of results. The performance results reported above are computed with improved evaluations and thus are not directly comparable with performance results found in previous Gemini model cards.

We expect variation in our automated safety evaluations results, which is why we review flagged content to check for egregious or dangerous material. Our manual review confirmed losses were overwhelmingly either a) false positives or b) not egregious.

Human Red Teaming Results

We conduct manual red teaming by specialist teams who sit outside of the model development team in Google’s Trust & Safety organization. High-level findings are fed back to the model team. For child safety evaluations, Gemini 4 Argon satisfied required launch thresholds, which were developed by expert teams to protect children online and meet Google’s commitments to child safety across our models and Google products. For content safety policies generally, including child safety, we saw similar or improved safety performance compared to Gemini 3.1 Pro. Additionally, the scope of red teaming covered potential issues outside of our strict policies, compared performance to Gemini 3.1 Pro, and found no egregious concerns.

Frontier Safety Assessment

In line with our Frontier Safety Framework (April 2026), we conducted critical capability assessments of Gemini 4 Argon to understand the inherent risk of the model, that is, whether a model may, without appropriate mitigations, contribute to the risks identified across the four risk domains. The results of the critical capability assessments, as measured against the relevant T/CCLs set out in our FSF, are summarized in the table below.

As outlined in our FSF, models reaching TCLs or CCLs will be subject to residual risk assessments to evaluate whether the deployment mitigations have brought the residual risk to an acceptable level. For models reaching CCLs, the residual risk assessment will be informed by a supplemental safety case outlining the mitigations.

For Gemini 4 Argon, drawing on internal and external results, the inherent capability thresholds that the model may have reached and our response are as follows:

• CBRN CCL: We cannot rule out that Gemini 4 Argon has reached this CCL for biological information risk, and we therefore developed a safety case for its biorisk capabilities that we assess lower risk to below the biological CCL level.
• Cybersecurity CCL: We cannot rule out that Gemini 4 Argon has reached this CCL, and we therefore developed a safety case for its cyberoffensive capabilities that we assess lower risk to below the cybersecurity CCL level.
• Stealth and Situational Awareness TCL: We cannot rule out that Gemini 4 Argon has reached this TCL, and we therefore performed a residual risk assessment for this risk that we assess lower risk to below the TCL level.

We continually work to improve the coverage and robustness of our frontier safety safeguards. The relevant safety cases and residual risk assessments outline the mitigations we have put in place.

Risks and Mitigations

For more information about the risks and mitigations, see the Gemini 4 Argon Safety & Security Report (coming soon on October 13).