Surveillance firm proposes "Border GPT"
In a webinar attended this morning by the author, Babel Street VP of Global Solutions Declan Trezise proposed border agents use large language models to summarize surveillance data on travelers.
Yesterday the Virginia-based company Babel Street announced its “Insight GPT” tool for simplifying interactions with its vast troves of internet surveillance data. Today, Babel’s VP of Global Solutions — a former theoretical physicist named Declan Trezise — floated a version for use by border agents which he christened “Border GPT”.
In contrast to Trezise, the other two panel members previously worked in US or UK border enforcement and had become cautious about law enforcement technology adoption as a result of direct experience. Trezise nevertheless persisted in promoting the potential of Babel Street’s artificial intelligence to simplify border crossings:
“I love the idea that there could be a model — a risk model — that could be built. A large data risk model. You know, a Border GPT.”
Before giving the system its punchy name, Trezise had laid out the motivation for such a system, as well as the combination of surveillance systems which might fuel it:
“A [traveler’s] risk can be assessed quite quickly if you have good biometrics. But in the absense of that, you may want to pull together things like publicly available information. You want to get a watch list that can be updated as quickly as possible with as much accuracy as possible. The border guard wants to know everything they can about a person before that person is in front of them. That’s where we’re going to start to focus some emerging technologies. We’re on the cusp of a revolution in terms of AI.”
Airports have increasingly justified facial recognition biometrics as a means of speeding up the identification process for large crowds of travelers. And, in the context of Babel Street, the term of art ‘publicly available information’ can be understood to include the social media surveillance capabilities of their Babel X software and the cellphone location-tracking data from their Locate X product.
Trezise even directly suggested that border agents monitor travelers’ vehicles as they approach:
“I think that the data could be collected as somebody travels, as somebody moves towards the port of entry. As they move towards the land border. We can look at their car, we can cross-reference that with publicly available information and understand are they on a watchlist. In the example where the 99.99% of people that are innocent, I’d like the idea that we could facility them to cross a border without the need to see another person before they get onto a plane or get onto their train or boat.”
Trezise also asked: “Could there be large models that exist from previous years of decision-making that border guards made, for example?”
Michael Ivahnenko, a former senior advisor to U.S. Customs and Border Protection who “managed CBP’s open source capabilities” and now ostensibly works at the border security firm Biometrica, was not completely opposed to the idea of incorporating large language models into the workflows of border agents. He acknowledged that “Most border guards are law enforcement; they’re not trained formal analysts”, and argued that “GPT and LLM capabilities” could help agents “to delve down and slice-and-dice data”, such as in summarizing cocaine seizures across the U.S.-Mexico border.
But Ivahnenko also paraphrased from the January 2022 report on Commercially Available Information which was recently declassified by the U.S. Office of the Director of National Intelligence:
“‘Commercial Available Information is increasingly powerful for intelligence and increasingly sensitive for individual privacy and civil liberties’. So that kind of sums it up. Whether you’re law enforcement, intel community, etc….The plethora of information and how…you utilize it to combat multiple threats is going to be a challenge given that there are serious first amendment and civil rights, civil liberties [concerns].”
The third panelist, former interim Director General of the U.K. Border Force Anthony John Smith, noted that he had been in charge of border security for London’s 2012 Olympics because of the theory that “lightning wouldn’t strike three times” after his previous positions as head of UK ports of entry during the July 7, 2005 bombings of London’s transit systems and as head of Canadian ports during 9/11.
Smith claimed that he has since worked with numerous companies which “wasted an awful lot of money on something that’s…not going to help the border officer at all” and emphasized the importance of tech companies such as Babel engaging with border enforcement agencies through matchmaking nonprofits such as the one he now chairs, IBMATA:
“Quite often these relationships get very very difficult, because quite often salesmen are trying to sell something, and well the buyers in the government haven’t really got the inclination or the money to buy it. That’s one of the reasons we set up IBMATA”.
Beyond U.S. Customs and Border Enforcement and the U.K. Border Force, another notable member of the International Border Management and Technology Association is Basis Technologies, where both Trezise and Babel’s Chief of Innovation, Catherine Havasi, worked until the beginning of this year. Both executives moved to Babel in January as part of the acquisition of BasisTech’s “Rosette” natural language processing capabilities, which now underpin the surveillance company’s rebranding as an artificial intelligence provider.
(The embattled facial recognition firm Paravision, which was formerly known as Everalbum, is also listed as a member of IBMATA. As is NEOM, the futuristic city brain-child of Saudi Arabian Crown Prince Mohammed bin Salman.)
When I asked the panel about border agent usage of social media surveillance, facial recognition, and Babel Street’s cellphone location-tracking product, none would answer in any meaningful detail. Ivahnenko gave the rote response that: “CBP utilized several commercial capabilities, uh, utilizing the publicly available, commercially available information to assess risk. There’s lot of open source information available on that subject. So I definitely would, uh, throw that out to do some research, uh, and just leave it as that.”
Despite Ivahnenko, as well as the rest of the panel, having dodged my question, he had earlier advertised his interest in “robust” discussions about surveillance techniques:
“You’re gonna be damned if you do and damned if you don’t…You want to have the public’s trust and you don’t want to get into those situations where you’re like ‘Okay, you caught me’, and then you degrade everyone’s ability across the board. For me, I appreciate the debate, and I look for robust, you know, discussions as it relates to it, but understand as an agency that has a mission set to protect the country, there is available information that we know should be utilized.”
Earlier in the conversation, Babel VP Declan Trezise noted that “borders are headline making. And [that’s] something that we all have to experience from time to time”.