TwelveLabs and Mantis Solutions —Video Brand Safety and Compliance That Sees What Text-Based Tools Can't

About Mantis

Reach PLC is the UK's largest commercial news publisher, home to the Daily Mirror, Daily Express, and 100+ regional brands. Mantis Solutions is Reach's technology division providing brand safety, contextual targeting, and data infrastructure for digital advertising.

The Stakes Just Got Higher

Video is the fastest-growing content format on every news platform and the hardest to govern. Publishers are producing more of it, syndicating more of it, and monetizing more of it through advertising than ever before. That growth isn't slowing down. As AI accelerates content creation across the industry, the volume of video that needs to pass through brand safety and compliance pipelines will compound every year.

The organizations that win aren't the ones with the most content. They're the ones whose brand safety and compliance infrastructure can keep pace with it automatically, accurately, and at scale.

Mantis Solutions, Reach PLC's technology arm, saw this problem clearly. They built the solution before it became a crisis.

The Problem Every Publisher Recognizes

Reach PLC is the UK's largest commercial news publisher: the Daily Mirror, the Daily Express, and over 100 regional news brands. Every day, thousands of articles go live across their portfolio. Many carry embedded video. Every piece of that video, if it runs advertising, has to meet brand safety standards.

A video about a natural disaster might be essential journalism. It might also be the wrong context for a luxury travel brand. A segment covering a violent crime is editorially important and commercially sensitive at the same time. The editorial team makes these calls instinctively. The problem is scale: no team is large enough to watch every video before every ad placement.

Before TwelveLabs, Mantis's brand safety and compliance pipeline was built on text. Articles were analyzed using Natural Language Understanding. Ads were placed or withheld based on what the article said. But video was a blind spot. Manual review couldn't keep pace. Existing tools could label individual frames. They could tell you a frame containing fire, but they couldn't tell you whether the video was covering a wildfire evacuation or showing a building burning down in a crime report. Mantis didn't need a frame classifier. They needed a model that could watch a video, follow the narrative, and produce a structured brand safety determination with the reasoning to support it.

Meeting the Brand Safety and Compliance Bar

Callum McAdam, Senior Technical Product Manager at Mantis Solutions, led the search for a solution that could meet the brand safety bar their advertising clients required at the speed and scale their publishing operation demanded.

"We'd been trying to close the video brand safety and compliance gap for a long time. Frame classifiers and manual review workflows struggled to interpret what was actually happening in the video, whereas Pegasus can actually understand that context. The fact that it explains its reasoning, not just its score, is what makes it workable for our brand safety and compliance team."

— Callum McAdam, Sr. Technical Product Manager, Mantis Solutions

The Automated Brand Safety and Compliance Engine: Pegasus on Amazon Bedrock

Mantis deployed TwelveLabs Pegasus on Amazon Bedrock to build a fully automated video brand safety and compliance workflow that slots directly into their existing brand safety infrastructure without a new pipeline or a manual handoff.

Pegasus processes the full multimodal signal — visual, audio, and textual — as an integrated whole. It doesn't sample frames and guess. It tracks events, context, and intent across the entire duration of a video.

A brand safety and compliance analyst doesn't get a list of flagged timestamps to scrub through. They get a structured assessment with a pass/fail determination, a safety score, and the reasoning that produced it, plus thumbnail previews at flagged moments. No one has to re-watch an entire video to understand why it was flagged.

Three Capability Pillars

Specificity and Structured Reasoning

Mantis didn't run a demo. They ran Pegasus against their actual content: the edge cases that consistently broke other tools — long-form news segments, emotionally ambiguous footage, videos where the visuals and the narration tell different stories.

The results delivered a level of specificity that prior approaches couldn't reach. Moving beyond simple object detection, the model provided structured reasoning, detailing exactly what was happening in the video, why it met or failed brand safety thresholds, and the confidence score behind every determination.

That accuracy and structured output format were the reasons the engagement moved to production.

The engagement began in Q4 2025. Mantis validated Pegasus against 70–80+ videos during the POC with consistently positive results on brand safety accuracy. The deal closed and moved to production in March 2026 — approximately six months from first engagement to live deployment.

Delivering Speed and Business Value

Mantis' operations team no longer watches footage to make brand safety and compliance calls. Pegasus watches it for them and explains its reasoning.

A Repeatable Pattern for the Industry

The engagement with Mantis Solutions represents a repeatable pattern for the industry. Any organization monetizing video through advertising faces the same critical gap: text-based brand safety tools cannot see the content they govern. Pegasus on Amazon Bedrock closes this gap with a production-ready solution that integrates seamlessly into existing enterprise infrastructure.

• Mantis contributes labeled production data to improve Pegasus for brand safety and compliance use cases.

• Real-world data and accountability help advance video understanding models.

• Automation allows reviewers to focus on nuanced edge cases requiring human judgment.

• The system scales alongside growing video volumes without increasing manual review.

• AI-driven brand safety creates a structural advantage over manual review workflows.

• Publishers relying on frame-level classifiers risk falling behind as video content continues to grow.

Technical Architecture