“The nature of threats facing public spaces and critical infrastructure has changed,” said Harry Mead, CEO and co-founder of London-based AI security startup Augur, referring to the rise of so-called “grey zone” attacks in Europe.
These threats, often coordinated by hostile actors to try to destabilize democratic nations, have targeted electricity grids, satellite communications and transport infrastructure, and Mead is using AI to fight back.
“Incidents are faster, more dispersed and often designed to exploit gaps. Augur exists to close those gaps,” he said.
The startup uses AI to make sense of real-time video data from sensors like CCTV camera networks, helping security operators flag suspicious activity to prevent threats before they happen and track suspects more effectively.
But processing high volumes of video and audio data in real time, in scenarios where every second counts, creates complex infrastructure questions. These companies need high-performance compute that complies with tight regulations concerning privacy and surveillance, as the era of AI crime fighting creates new challenges for the cloud industry.
Public safety meets privacy
Augur’s AI technology is designed to protect against threats to critical infrastructure, as well as terror attacks and other criminal activity in large public spaces like stadiums or shopping malls, combining data from legacy hardware like CCTV camera networks with AI vision models.
From its UK headquarters, Augur set out to provide European countries with public safety technology that is compatible with regional laws and values around the right to privacy, and made the decision not to use facial recognition to track individuals.
“Our mission at Augur is to make public spaces and critical infrastructure safer without compromising privacy or civil liberties,” said Stefan Kopieczek, Augur co-founder and head of engineering, adding that steering away from facial recognition also makes tracking people in busy spaces more effective.
Augur co-founders, Imran Lone (L), Harry Mead (C) and Stefan Kopieczek (R). Credit: Augur
“This approach is much more robust in the real world where image quality is variable and you can’t rely on getting regular face captures. So in this case, the right thing to do is also the best solution in engineering terms.”
Augur has achieved this by rethinking how person tracking works: instead of simply drawing a box around a detected individual in an image, it uses the position and orientation of cameras to infer where the person is in 3D space.
“Our tracker combines those spatial features with visual features, and the combination of the two means we can be very confident in a match, even when the person isn’t consistently visible in the camera frame,” said Kopieczek.
Augur came out of stealth mode in March 2026, and said it has already signed contracts with football stadiums, retail centers, transport hubs, power plants, data centers and military sites, that see the potential of its AI to help catch suspects and prevent threats before they happen.
A step ahead of the criminals
Syntelligence, a London-based joint venture established by major telecom providers, is tackling a different kind of threat: the scam call.
“Scam calls are one of the key problems that telecom providers have been suffering from for the past two decades,” said Prateek Choudhary, CEO of Syntelligence. “Even though it's such a persistent problem, and pretty much impacts everyone on the planet who has a mobile phone, it's not really solved and it's actually becoming progressively worse.”
Syntelligence uses AI to try to prevent scam calls before they happen, based on metadata like whether a phone number has been used to call multiple new numbers in a short space of time, and sends a warning to the recipient. It then gives people the option to send the call to an AI agent that can assess whether it’s a scam before they take the call, or they can ask it to listen live.
Prateek Choudhary, CEO at Syntelligence. Credit: Syntelligence
“We are using speech to text, and then we get the transcription, and the LLM analyzes that transcription,” said Choudhary.
“It could be someone claiming they are calling from the police, saying you have an outstanding payment and you need to do it now, otherwise we are coming with a warrant. Essentially creating this artificial urgency for giving quick payment information. As soon as that is happening, our model will understand that it's evolving into a scam, and it will give you live alerts.”
The threat of scam calls has become even greater due to the rise of audio deepfakes, which can make malicious attempts harder than ever to spot.
“This is essentially someone calling you pretending to be someone else, but using exactly the same voice. We have seen cases like that happening already, and it will probably become more common, so we have to fight this,” said Choudhary.
Crime prevention in the cloud
“Stadiums range from high tens to low hundreds of cameras. We preprocess the video streams on premise in order to reduce the network bandwidth, but it can still be as high as gigabits of data per second for a large site,” said Kopieczek.
Alongside the ability to scale capacity and offer low latency, the companies have distinct needs from the cloud industry.
For Syntelligence, some of the hardware required to power its AI solution will come from telecom providers that run their own data centers and, in cases where that is combined with additional cloud compute, it will require seamless integration.
Augur’s biggest requirements from cloud providers are data sovereignty, governance, access to the latest GPU hardware and capacity to scale compute needs in response to customer demand.
“In some cases, the data needs to stay within the originating country, whether that is the UK or an allied nation,” said Kopieczek.
“We’re also keen to question the assumptions of the industry and push the envelope of what’s possible. We’ve had some great conversations with forward-looking providers who are willing to be partners on that journey when it comes to frontiers like large-scale fine-tuning of vision-based LLMs and experimenting with new hardware.”
