The inference machine
Surveillance technology is not just watching anymore – it’s judging
Welcome, readers! Dropping into your inboxes late this week with a fresh set of stories on digital censorship and surveillance around the globe.
This week, I learned about the UK government’s plans to use AI to estimate the age of immigrants and it made me think again about how surveillance technology is changing in this age of artificial intelligence. For decades, surveillance tech has been about gathering data and storing it for occasional review and analysis by people, if there was a need.
Now, the very nature of this technology is changing: AI can ingest and analyze massive amounts of data in real time, making inferences, suggesting conclusions and offering possible actions. A camera doesn’t just take your picture or record a video of you walking down the street. It can also potentially guess your age, your mood or your health conditions.
It can suggest that you’re walking in a hurry, or that you are stressed, confused, not feeling well, or intoxicated. It can notice you’re not on your usual route at this time of the day, make a guess about what you’re carrying in your hands, register signals and patterns that turn a simple picture of you walking down the street into a rich data stream about your life.
This imaginary scenario is not the reality for most places on Earth so far – but we know that China, for example, is already working on predictive policing technologies that are designed to do exactly this: make inferences about people’s behaviour and take action in real time, both in the physical world and in the digital realm.
For democratic societies, these kinds of scenarios can be prevented if people decide that a certain amount of surveillance is too much. For less democratic ones, whatever promises the most control over the masses will be implemented with little resistance.
But for all of us, knowing what is happening behind the shiny technological facade is important – even though it can be anxiety-inducing and frustrating at times. So, without further ado, let’s get into this week’s stories!
As always, I hope that you’re enjoying this newsletter and if so, that you would take a moment to share it with friends – that “new subscriber” alert in my inbox makes my day brighter every time! Just kidding (not kidding).
Biometrics briefing
A man in Utah was wrongfully accused of damaging an Uber car because of a facial recognition mistake. – Idaho News
Kansas City, Missouri, is planning to equip some public buses with facial recognition-enabled cameras to identify banned passengers or missing persons. – Trail Gazette
Australian police will start using real-time facial recognition cameras in the state of Western Australia. – The West Australian
Not just any facial recognition
Meta is testing a military-grade facial recognition technology for its smart glasses, Wired reports. The company has licensed technology from Rank One Computing (ROC) – a Denver-based company that works mostly with government agencies, especially the military.
Among ROC’s clients are police departments, the U.S. Marshals Service, the U.S. Special Operations Command, and the Naval Criminal Investigative Service. The company leadership includes FBI, CIA and Pentagon veterans, including its CEO, B. Scott Swann, who previously oversaw the FBI’s work with biometric databases.
The U.S. Marshals Service is using ROC’s facial recognition to identify prisoners during transport. According to the documents reviewed by Wired, Meta acquired a license for using Rank One’s face recognition and liveness detection, which allows checking whether a face the camera “sees” is a real person rather than a mask or a photo. Also, ROC’s technology is used by schools in West Virginia to screen people entering schools against the state sex offender registry.
Wired had previously reported that Meta had already deployed the facial recognition code for its smart glasses in its AI mobile app, although the feature wasn’t live. The company removed the code after Wired‘s publication, but fragments of the ROC integration code are still in the app.
ICE watchers’ Signal chats
The Intercept analyzed the indictment against 15 Minnesotans who observed the ICE raids in the Twin Cities this winter. The activists are facing charges like “conspiracy to impede or injure an officer,” “solicitation to commit a crime of violence,” and “destruction of government property.”
According to the indictment, a lot of information on the defendants’ activities was derived from the Signal chats where they coordinated, The Intercept notes, however, that the prosecutors also cited data on Signal voice calls, which could indicate that the investigators had obtained access to the activists’ phones.
The FBI started an investigation into Minnesota anti-ICE activists in January after influencer Cam Higby infiltrated their Signal chats. Higby also boasted about infiltrating New Jersey anti-ICE activists’ chat groups recently.
There is no indicator of Signal itself having been hacked or otherwise compromised so far. The app has become the messenger of choice for journalists and activists around the world due to its extensive privacy features. In April, Apple fixed a bug that previously allowed retrieving deleted Signal messages using police forensic tools if those messages had been saved in the iPhone’s notification database.
Tax data for ICE
Last year, a U.S. court prohibited the IRS from sharing taxpayers’ data with the Immigration and Customs Enforcement (ICE) – the collaboration was supposed to help ICE identify undocumented immigrants by their individual taxpayer identification numbers, or ITINs. But the agency has found a loophole – now it’s buying the data from a private broker, 404 Media reports.
ICE has signed a $10 million contract with Thundercut Technology for “ITIN data subscription and analytics for HSI agents in fraud investigations.” ITINs are given to people who cannot get a Social Security Number.
This is just another piece of a larger trend: instead of obtaining personal data through warrants and subpoenas, U.S. government agencies are buying it from private data brokers. For example, the FBI and ICE have been buying cell phone location data from commercial providers. The trend is not new or specific to the current administration: back in 2021, The Washington Post reported on ICE purchasing data on U.S. utility users from Thomson Reuters’ CLEAR database.
Let a robot guess your age
The UK is planning to introduce facial age estimation (FAE) for asylum seekers, Wired reports – FAE is an AI-based technology that is supposed to estimate the age of a person based on a face scan. Many immigrants who come to the UK seeking asylum lack reliable documents, so officers at the border have to verify their age by visual assessment. To help them, the Home Office is betting on AI, even though the technology has proven unreliable, Wired and Lighthouse Reports found.
According to a leaked document from the Home Office, the agency tested seven facial age estimation algorithms from an unnamed provider last year. The technology worked much worse on Sub-Saharan Africans than on any other group, according to the report, which sounds like an issue as people from Sub-Sarahan Africa are the largest group of immigrants coming to the UK by water.
The system performed especially badly on female African faces: the results were off by 4.6 years on average. The report also notes that experiencing trauma and a stressful trip – which is the case for many asylum seekers – can make people “temporarily age,” affecting the accuracy of an automatic estimate.
Facial recognition technology is notorious for its racial bias as well. However, the Home Office has apparently decided to fully focus on AI and disbanded a scientific committee designed to study other age estimation methods, Wired writes.
And it looks like the technology is here to stay no matter what, especially as more countries are introducing age restrictions for social media and there is a growing need to verify users’ age online. Australia banned teenagers under 16 from social networks in December, Indonesia did it in March, and Malaysia is implementing the same policy now. In Europe, the UK, France, Spain, Greece, Austria and Denmark are currently considering similar measures – read this overview by The New York Times for more details.
In May, Meta announced that it would start using AI age estimation for the profile pictures of users in the EU and Brazil. There are also local initiatives like a Louisiana bill suggesting biometric age verification for admitting people into bars.
France doesn’t trust Palantir
Palantir, a controversial tech giant providing the U.S. government with surveillance software, is getting some rejections in Europe. The company has gained notoriety for its extensive collaboration with the U.S. government, technological support for the current anti-immigration crackdown, and its closeness to the Trump administration.
This week, The Telegraph reported that France’s domestic intelligence agency, the General Directorate for Internal Security (DGSI), would stop using Palantir’s software it had earlier purchased. The decision was prompted by the fact that another powerful American corporation, Anthropic, paused access to its most advanced models, Fable and Mythos, due to pressure from the White House.
In response, French politicians decided the country could not depend on U.S.-made software, access to which may be switched off overnight because the Trump administration feels like it. Prime minister Sebastien Lecornu said France would invest €655m in the homegrown artificial intelligence. The transition might not be easy, as domestic alternatives are currently considered not to be on par with Palantir, The Telegraph writes, but the French data analytics firm ChapsVision is a likely replacement.
Palantir is getting some pushback in the UK, too. The country’s parliament members are urging the government to end Palantir’s contract with the National Health Service for the creation of a national patient database, and London’s mayor Sadiq Khan recently blocked the firm’s contract with the Metropolitan Police.
No Telegram in India
India has blocked Telegram until June 22 to prevent cheating during the national admission tests for medical schools, Reuters reports. According to the news agency’s sources, Google and Apple have received a government order to temporarily remove Telegram from their app stores and will comply.
The Ministry of Education’s National Testing Agency said the ban was “in response to the organised use of the platform by cheating rackets to defraud candidates.” The test is scheduled for June 21.
In May, the exam’s questions were allegedly leaked and sold on Telegram channels. The results of the exam were canceled for 2.3 million students, spurring protests across the country.
Telegram’s founder Pavel Durov criticized the decision in a blog post, saying the company “removed hundreds of channels sharing leaked exam materials and related scams in India,” but the leaks appeared on other messengers as well. “This punishes 150 million ordinary Telegram users in India – not the insiders who leaked the exam materials,” Durov added.
***
And that’s all from me for this week, guys.
Stay vigilant!
Anna