Milestone launches AI tools for security video teams

Milestone Systems has announced a set of artificial intelligence tools for security operations: AI Search, XProtect Video Summarization and Video Anonymization.

The software is intended to help security teams review footage, document incidents and redact personal data in video. Milestone framed the launch as a response to operators' concerns about accuracy, false positives and the use of generative AI in settings where decisions can carry serious consequences.

Security teams have been slower than some sectors to adopt generative AI, as many still rely on manual review of recordings, written reporting and hand-led anonymisation of footage. Milestone also pointed to growing pressure from privacy and AI regulation, as organisations handling large volumes of video face questions about how to extract value from those archives while staying within regulatory limits.

Product rollout

AI Search is designed to let investigators query video using natural language instead of filters or manual scrubbing. Milestone said an operator could describe a vehicle involved in a traffic collision and retrieve matching footage from multiple cameras in seconds.

XProtect Video Summarization generates text descriptions of footage and can be used to produce incident reports. The tool can run automatically on alerts, on a schedule or on request through XProtect's rules engine.

Video Anonymization redacts faces and number plates before footage is exported or shared. It is aimed at organisations that need to provide video to outside authorities or move it between departments without exposing personal identities.

Video Anonymization will be available in Arcules and XProtect. AI Search is scheduled to be available by the end of 2026, while an on-premises version for secure and air-gapped environments will be offered through BriefCam.

Data focus

The new tools are built on vision language models trained on security-specific video rather than general-purpose models adapted from other fields. Milestone said the models were fine-tuned on use-case-specific datasets drawn initially from traffic scenarios in Europe and the US.

According to the company, the underlying video library is assembled through fair-exchange agreements with customers, anonymised using brighter AI technology and tagged with more than 300,000 categorisations using NVIDIA technology. It said the resulting dataset is traceable and structured to preserve privacy.

Milestone said the data is used to fine-tune NVIDIA Cosmos Reason foundation models. In its own testing, it reported a 19.4% improvement in flow and direction correctness and a 4.4% improvement in alert verification accuracy over base foundation models for a traffic-focused vision language model.

The city of Genoa is the latest contributor to the library, adding video data intended to broaden the dataset's geographic spread. Milestone said the data foundation is designed to align with the EU AI Act and GDPR, two regulations shaping how organisations can use video and AI in Europe.

Trust question

Andrew Burnett outlined the rationale behind the launch in a statement accompanying the announcement.

"How can we build on our decades of experience in video management to prepare our customers for the next generation of artificial intelligence? The answer is trust. Trust that the data is compliant, trust that the AI is accurate, and trust that these tools will genuinely make operators' jobs easier. That's what we're delivering today," said Andrew Burnett, Chief Technology Officer at Milestone Systems.

Milestone employs more than 1,500 people and sells video technology used in sectors including manufacturing, airports, law enforcement, retail and traffic management. Its portfolio includes XProtect video management software, BriefCam analytics and Arcules cloud video services, alongside work on anonymised video datasets for AI training.

The launch places the company more directly in a market where suppliers are trying to apply generative AI to surveillance and operations software without increasing regulatory exposure. Milestone's pitch rests on two issues that have become central to buyers: whether the models can produce dependable outputs in real security workflows, and whether the data used to train and operate them can withstand scrutiny under privacy law.

Milestone said its video training library is responsibly sourced, fully anonymised and traceable, with brighter AI technology used in its internal data pipeline as well as in the customer-facing anonymisation product.