Last updated: November 18, 2026
What does "Gemini security" mean when the model generates video? Google made Gemini Omni 1.2 Flash generally available on November 5, 2026, and Gemini security for video differs from text in four ways: synthetic media misuse, runaway spend, reference-file exposure and missing provenance. This checklist gives security teams ten controls to apply before production.
TL;DR: Key Takeaways
- Gemini Omni 1.2 Flash reached general availability on November 5, 2026, with clips of up to 15 seconds and a published rate of $0.07 per second at 720p.
- Google states it does not train or fine-tune models on customer data without prior permission, but zero data retention is not automatic and caching can hold data for up to 24 hours.
- The FBI's 2025 report counted 22,364 AI-related complaints and $893 million in losses, so synthetic media is already a measurable fraud channel.
- EU AI Act Article 50 has applied since August 2, 2026: providers must mark synthetic output and deployers must disclose deepfakes.
- Shadow AI use rose from 15% to 45% of employees in one year, which makes the Gemini app, Flow and Vids a governance gap.
- A 15-second clip at 720p costs about $1.05, so 10,000 clips cost about $10,500 (our calculation).
Gemini security at a glance
| Control area | Main risk | What Google provides | What your team owns |
|---|---|---|---|
| Data handling | Prompts and reference files kept longer than expected | No training on customer data without permission; caching limited to 24 hours | Logging settings, retention policy, reference-file rules |
| Cost | Runaway generation bills | Provisioned Throughput for capacity | Budgets per team, rate limits, alerts |
| Provenance | Unlabeled synthetic video | SynthID watermark on outputs | Verification workflow, disclosure labels |
| Access | Shared keys, unsanctioned tools | Enterprise plans on Google Cloud | Key hygiene, one approved access path |
| Misuse | Impersonation and fraud | Usage policies | Review gates, call-back procedures |
What is Gemini Omni 1.2 Flash?
Gemini Omni 1.2 Flash is Google DeepMind's video generation and editing model, available since November 5, 2026 through the Gemini API, Google AI Studio and Google Cloud. It adds multi-turn editing, reference-guided generation and lip-synced speech to clips of up to 15 seconds. Google's launch post is the source for every specification below.
| Specification | Omni 1.2 Flash |
|---|---|
| General availability | November 5, 2026 |
| Clip length | Up to 15 seconds (10 seconds in Omni 1.1) |
| Draft resolutions | 360p or 720p |
| Final upscale | 1080p or 4K |
| Published price | $0.07 per second at 720p, 30% lower than Omni 1.1 |
| Inputs | Text, reference images, motion clips up to 3 seconds, audio cues |
| Where it runs | Gemini API, Google AI Studio, Google Cloud, Flow, Google Vids, Google Ads, Gemini app |
| Provenance | Imperceptible SynthID watermark |
How we built this checklist
We read Google's launch post and Google Cloud's data governance documentation, then mapped each feature to a control using OWASP's 2026 LLM list, FBI complaint data and EU AI Act guidance. Google's material does not describe video-specific retention, so those points need confirmation from your account team.
Gemini security risks that are specific to video generation
Video generation adds risks that text-only reviews miss. The first is impersonation. According to the FBI (2026), its Internet Crime Complaint Center logged 22,364 AI-related complaints and $893 million in losses for 2025, a likely undercount. Realistic speech at a low per-second price lowers the cost of synthetic footage for legitimate teams and attackers alike.
The second is untrusted input. According to OWASP (2026), prompt injection keeps the top spot in the LLM Top 10 and now covers cross-modal attacks hidden in images or audio. Omni accepts reference images, motion clips and audio cues, so a poisoned file chosen by an agent becomes an instruction channel.
The third is spend. Unbounded Consumption is a named category in the 2026 list, and video is billed per second.
| Volume of 15-second clips at 720p | Cost at $0.07 per second |
|---|---|
| 100 clips | $105 |
| 1,000 clips | $1,050 |
| 10,000 clips | $10,500 |
These figures are our calculation from the published 720p rate. The launch post does not state 1080p or 4K prices.
The fourth is governance drift. According to Verizon's 2026 DBIR, shadow AI use rose from 15% to 45% of employees in a year. Omni 1.2 Flash is also in the Gemini app, Flow and Vids, so staff can create video outside any approved path.
Gemini security checklist for Omni 1.2 Flash
Work through these ten controls in order. The first four decide whether the rest can be enforced.
1. Choose one approved access path
Decide whether teams call Omni through the Gemini API, Google Cloud or a Gemini business plan, and block the others. Each path has different terms, logging and access controls, so one route underpins every other control. Assign an owner.
2. Confirm what Google logs and keeps
Google Cloud's documentation says it will not train or fine-tune models on customer data without prior permission. It also says zero data retention is not automatic: caching can hold data for up to 24 hours, and abuse monitoring may log prompts for customers under the Google Cloud Platform Terms. The page lists no video-specific terms, so ask your account team how it applies to Omni.
3. Set spend limits before launch
Give each team a monthly budget with alerts, and cap clip length and resolution in your application code. Google's workflow drafts at 360p or 720p and upscales only approved clips, so enforce that order.
4. Treat every reference file as untrusted input
Reference images, motion clips and audio cues come from users or other systems. Check file type and size, strip unneeded metadata and review any reference an agent picks from untrusted sources.
5. Restrict real people and brand assets
Write a rule for footage of employees, executives, customers or public figures. Require written consent and a named approver, and keep likeness assets in a controlled library.
6. Keep SynthID and add your own labels
Omni outputs carry an imperceptible SynthID watermark. The SynthID Detector portal opened to the public on October 6, 2026, but it is not a general AI detector and covers content from four companies: Google, OpenAI, Nvidia and Kakao. A clean result does not prove a clip is real, so keep original files and add visible labels for external use.
7. Meet disclosure duties for synthetic video
If you publish Omni output to EU audiences, Article 50 of the AI Act applies, as explained below.
8. Require human review before publication
Route external video through marketing or legal review, checking frames for logos, faces and on-screen text, and record who approved each clip.
9. Log the requests you control
Record who requested what, when, with which model and at what cost, in your own gateway. Do this even with zero data retention at Google; the record supports incident response.
10. Rehearse a synthetic media incident
Finance and executive teams need a call-back procedure for requests that arrive by video, and a takedown contact for fake clips of your brand.
EU AI Act Article 50 and synthetic video
According to the European Commission (2026), Article 50 has applied since August 2, 2026. Providers of systems that generate synthetic video must mark outputs in a machine-readable format, and deployers must disclose deepfakes to viewers. Morgan Lewis reports that the Digital Omnibus moved one date: providers' marking duties for systems already on the EU market before August 2 now run to December 2, 2026. The Commission's Code of Practice is voluntary. Teams publishing Omni output in the EU are likely deployers, so label it and confirm your role with legal.
How NeuralTrust secures Gemini workloads
Google secures the model. The controls around who calls it, how often and with what input remain yours. Agent Gateway (TrustGate) sits between your applications and the model APIs they call, so teams share one governed route instead of loose API keys, with per-team usage limits, request logging for audit and prompt inspection.
Agent Runtime Security (TrustGuard) adds runtime checks when an agent generates video as one step in a larger workflow, where an injected instruction does the most damage. NeuralTrust, a Barcelona-based AI agent security company, is recognized in six Gartner Hype Cycles for 2026, including AI Runtime Defense. See the AI Gateway and agent runtime security.
Which Gemini access path should you choose?
| Access path | Best for | What to check | Main watch-out |
|---|---|---|---|
| Gemini API and Google AI Studio | Prototypes and product teams | API key handling, billing limits | Key sprawl across teams |
| Google Cloud | Production workloads | Provisioned Throughput, data governance settings | Logging and caching defaults |
| Gemini for business | Workforce use in managed accounts | Admin controls for video | Confirm what admins can restrict |
| Gemini app, Flow and Vids | Individual creators and marketing | Who holds Ultra or Workspace seats | Easy to use off-policy |
Conclusion
Gemini security for Omni 1.2 Flash depends less on the model than on the route around it. Google documents no training on customer data and a SynthID watermark, but video retention is not spelled out, spend is metered per second and EU disclosure duties already apply. Fix the access path and the budget cap first, then add provenance and review.
Secure Gemini Workloads in Production with NeuralTrust
Put a governed gateway between your teams and Gemini.
Related Comparisons
- Google Gemini 4 Argon Release: Benchmarks, Pricing and Security
- Gemini 4 Carbon: Benchmarks, Pricing & Security (2026)
- GPT-6 Sol vs Luna: Security, Cost and Which to Deploy
FAQs about Gemini security
1. What is Gemini Omni 1.2 Flash?
Gemini Omni 1.2 Flash is Google DeepMind's video generation and editing model. It became generally available on November 5, 2026 and creates clips of up to 15 seconds with speech, reference-guided generation and multi-turn editing, through the Gemini API, Google AI Studio, Google Cloud, Flow, Google Vids, Google Ads and the Gemini app.
2. Does Google train on data sent to Gemini Omni?
Google Cloud's documentation says it will not train or fine-tune models on customer data without prior permission, but it lists no video-specific terms. Confirm with your account team how it applies to Omni, and review logging and caching settings, because zero data retention is not automatic.
3. How much does Gemini Omni 1.2 Flash cost?
Google's launch post publishes $0.07 per second at 720p, 30% lower than Omni 1.1. A 15-second clip at that rate costs about $1.05 and a full minute about $4.20 (our calculation). Check the pricing page for 1080p and 4K upscales.
4. Can Gemini Omni output be detected as AI-generated?
Omni outputs carry an imperceptible SynthID watermark, and the SynthID Detector portal is open to the public. It covers content from Google, OpenAI, Nvidia and Kakao and is not a general AI detector, so a negative result does not prove a video is authentic.
5. Do EU AI Act rules apply to Gemini video?
Article 50 has applied since August 2, 2026. Providers must mark synthetic output in a machine-readable format, and deployers must disclose deepfakes to viewers. If you publish Omni videos to EU audiences you are likely a deployer, so add visible labels and confirm your role with legal.
6. How do I stop employees from using Gemini video tools outside approved channels?
Pick one approved access path, block the others at the network and account level, and give teams a governed gateway with budgets. Verizon's 2026 DBIR found shadow AI use rose from 15% to 45% in a year, so policy alone will not hold.
About the Author
Roger Howroyd is Head of Global SEO and AI at NeuralTrust, where he leads the company's search strategy across SEO, AEO, GEO, and LLM optimization. He specializes in AI-powered search, content strategy, and SEM. Connect on LinkedIn.
NeuralTrust is the leading platform for securing and scaling AI agents. Named a Pioneer in the Gartner Emerging Market Quadrant for AI Application Security 2026, recognized across six Gartner Hype Cycle reports in 2026, and featured in the Gartner Market Guide for Guardian Agents 2026, the Gartner Market Guide for AI Gateways 2025 and the KuppingerCole Leadership Compass for Generative AI Defense 2025. Headquartered in Barcelona with offices in London and New York. ISO 27001 certified.
Sources
- Google, "Introducing Gemini Omni 1.2 Flash," November 5, 2026: https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-omni-1-2-flash/
- Google Cloud, "Gemini Enterprise Agent Platform and zero data retention," updated November 17, 2026: https://docs.cloud.google.com/vertex-ai/generative-ai/docs/data-governance
- Malwarebytes, "Americans lost nearly $900 million to AI-powered scams, FBI says," June 8, 2026: https://www.malwarebytes.com/blog/scams/2026/06/americans-lost-nearly-900-million-to-ai-powered-scams-fbi-says
- European Commission, "Guidelines on transparency obligations for providers and deployers of certain AI systems," updated August 6, 2026: https://digital-strategy.ec.europa.eu/en/policies/guidelines-ai-transparency-obligations
- Morgan Lewis, "EU AI Act's Transparency Rules: What Went Into Effect on 2 August?," August 12, 2026: https://www.morganlewis.com/blogs/sourcingatmorganlewis/2026/08/eu-ai-acts-transparency-rules-what-went-into-effect-on-2-august
- European Commission, "Code of Practice on Transparency of AI-generated Content," updated October 19, 2026: https://digital-strategy.ec.europa.eu/en/policies/code-practice-ai-generated-content
- PPC Land, "Google's SynthID Detector gains public access and covers 4 companies," October 7, 2026: https://ppc.land/googles-synthid-detector-gains-public-access-and-covers-4-companies/
- Help Net Security, "OWASP 2026 LLM Top 10," August 6, 2026: https://www.helpnetsecurity.com/2026/08/06/owasp-2026-llm-top-10-released/
- Verizon, "Vulnerability exploitation top breach entry point, 2026 industry-wide DBIR finds," May 19, 2026: https://www.verizon.com/about/news/breach-industry-wide-dbir-finds
)
)
)
)
)
)
)