Revealed: the secret algorithm that controls the lives of Serco’s immigration detainees
Imagine there’s a secret rating that dictates where you sleep and whether you are forced to wear handcuffs to a doctor’s appointment. And imagine that rating is based on incorrect information or unfair assumptions about the type of person you are.
In Australia’s immigration detention centres, each detainee is given security risk ratings decided by an algorithm – but they’re not even told it exists.
Developed by Serco, the company tasked with running Australia’s immigration detention network, the Security Risk Assessment Tool – or SRAT – is meant to determine whether someone is low, medium, high or extreme risk for escape or violence.
Related: Australia’s high court to hear two appeals over legality of re-detaining more than 100 non-citizens
Immigration insiders, advocates and detainees have told Guardian Australia the SRAT and similar tools used in Australia’s immigration system are “abusive”, “a blunt instrument” and “unscientific”. Multiple government reports have found that assessments can be littered with inaccuracies – with devastating consequences.
“One of the biggest problems with these tools is that there’s a complete lack of transparency … about how they work,” says Jathan Sadowski, a senior research fellow at Monash University who studies the use of risk analysis systems.
‘Judge, jury and executioner’
Over his 1,490 days in immigration detention, Nauroze Anees was not told about his security risk assessment. He only found out he had one after the Australian Human Rights Commission (AHRC) investigated his case.
When he saw what was included, he said it almost gave him “a heart attack”.
The document was riddled with errors and vague claims: a minor injury he sustained while playing football listed him as an “alleged offender”. Another incident was summarised as “detainees involved in assault”, but did not describe who was involved or even when or where it occurred. An incident described as a “minor disagreement” was logged as “abusive/aggressive behaviour”.
Anees arrived in Australia from Pakistan in 2007 on a student visa. When he stopped studying to care for his ex-partner who had serious medical needs, the visa was cancelled and he was then refused a partner visa on character grounds, after he committed criminal offences during a period of financial stress and homelessness.
He was released from detention in 2020 after a tribunal found he was “of good character” and had “contributed positively to the Australian community”, but his fight over the SRAT dragged on for three years after he made a privacy complaint against Serco to the Office of the Australian Information Commissioner.
In October 2023, the OAIC decided his privacy had been interfered with because Serco failed to ensure his personal information “was accurate, up-to-date and complete”, among other breaches of the Privacy Act. Serco was made to apologise and pay him $1,500.
But Anees feels the apology was insufficient and remains outraged by the power the tool had over him while he was completely unaware and unable to challenge its content.
“Serco is essentially the judge, jury and executioner,” he said.
What we know about the SRAT
Versions of these security risk assessments, described as “a series of mathematical calculations” by the home affairs department, have been in use since at least 2012.
The SRAT calculates a detainee’s “risk” for behaviours such as escape, demonstration and self-harm based, in part, on their pre-detention history and more than 30 incident types that may occur in detention, such as abusive or aggressive behaviour, assault, possessing contraband or the refusal of food or fluids.
Detainees are also rated for an overall placement and escort risk, which may determine how they are treated while being transported, such as whether they are placed in handcuffs, and where they are housed inside a detention centre.
Sarah* was an intelligence analyst on Christmas Island whose role included completing risk assessments for people in detention. She told Guardian Australia that each incident had a value associated with it, depending on the perceived level of seriousness. Incidents with the greatest weightings might be escape, riot, major disturbance and self-harm, for example. The department declined to reveal to Guardian Australia how the weighting system works.
It was relatively easy to accumulate incidents on your SRAT that added to your risk score, she said. An expression of frustration almost always came through as “disturbance, minor”, she said, or even as abuse or aggressive behaviour. “Something like that could basically tip someone’s rating into an almost irreparable rating.”
A 2019 AHRC report found “abusive/aggressive behaviour” was among the incident categories most commonly used by Serco analysts, and encompassed both the use of bad language and conduct that is physically aggressive but not assault.
These incidents were used to calculate a risk rating for “aggression/violence” even though the interaction may not have included any physical aggression or violence, the report said, “and may be an understandable product of long term immigration detention”.
One person currently detained at Villawood told Guardian Australia people are very aware of this risk. Some officers can be trusted not to record every minor outburst, he said. “Other officers, you have to be careful what you say around them, because they’ll write down anything. We know who those officers are.”
An assessment can also have a significant bearing on where a detainee is placed in the detention network, including in which compound inside a centre.
Multiple detainees told Guardian Australia that while people don’t explicitly know their risk assessment, it’s common understanding that compounds are sorted by risk level. “We aren’t made aware of what our security level is but it’s fairly obvious,” one said.
This means that inaccurate SRATS can have profoundly serious consequences, especially as the population in immigration detention has shifted from asylum seekers who arrived by boat to those who have overstayed visas or had visas cancelled on character grounds due to criminal charges.
Related: People seeking asylum in Australia shouldn’t have to trade danger for degradation | Frances Rush
In a 2023 AHRC report, a man who told officials he was 17 but was identified as an adult using now discredited age verification techniques when he arrived in Australia was identified as “high risk” in 2016, based on what the department said was his “propensity for violence”.
He had no criminal record in Australia and allegations about criminal behaviour overseas were deemed “vague, confusing and contradictory”. Likewise, most of the incidents recorded in his SRAT as “Abusive/Aggressive Behaviour” were verbal.
He was nonetheless placed in a high-risk facility and was assaulted by three other detainees, in what a court later called “a vicious beating”.
default
‘The veneer of objectivity’
A person familiar with immigration decision-making under the former Coalition government said tools like the SRAT were intended to provide “a fig leaf” to restrictive home affairs policies. Their purpose was to put a technological sheen over mandates that subjected everyone in immigration detention to extremely tight restrictions, despite their individual circumstances, he said.
“That was the edict: everyone despite their risk rating was treated as high risk.”
Sadowski said risk analysis tools are often built to achieve a particular outcome. “A major problem with these tools is that they are often built with a solution in mind,” he said. “You get the veneer of science and objectivity … but the social outcome was already preordained.”
The Australian Border Force commissioned a review by Griffith University, completed in 2019, on improving immigration detainee risk assessment after concerns were raised about the “validity” of assessments made using the SRAT.
In a report heavily redacted by home affairs, the researchers concluded that “while the SRAT has been the risk assessment tool of choice for the service providers contracted by the ABF, it is clear that the tool is not borne out of sound scientific research”.
As at 18 January 2024, at least 75% of immigration detainees were rated by Serco’s SRAT as high or extreme risk for both placement and escort, according to a document obtained by Guardian Australia under freedom of information laws. The proportion has largely stayed the same since at least 2020.
“There’s just no scientific basis for what they’re doing and no consistency,” the director principal of Human Rights for All, Alison Battisson, said.
According to sources who have worked in immigration detention and lawyers who represent detainees, there’s also a perverse incentive implicit in the SRAT system because Serco might be entitled to charge the government more for detainees rated high risk by its own system.
Security risk assessments for each detainee are required by Serco’s 2014 contract with the then-Department of Immigration and Border Protection, which also notes “the Service Provider is entitled to charge the Department additional fees for High Needs Detainees”. It says the needs and circumstances of individuals might include physical and mental health issues or “differing level of risk assessments”. The contract was set to expire 10 December 2023 but was extended by a year.
default
Riddled with inaccuracies
Adam*, who was in immigration detention and has since left Australia, shared several of his SRATs with Guardian Australia, which he obtained via FOI. He claims they contained inaccuracies that led to him wrongly being rated as high risk for escort, and therefore more likely to be restrained.
In one of his SRATs, in a box titled “escort notes” – detailing how he should be treated while travelling outside a detention centre – it was claimed he was “aware of wider Sydney region” and was “initially assessed as HIGH risk for escort”.
“I found [that] entirely laughable because Sydney is actually one of the capital cities in Australia that I’ve never ever been to,” he said.
Related: ‘We can be a role model’: the activism of Hazara women in regional Australia
His SRAT also notes that he has been “supporting disruptive work by refugee advocates” and “also advocates for detainees”.
“I’m not at all surprised by it, because this is kind of indicative of how the whole system is run,” Adam said. “My entire experience through the immigration detention system was that it’s a little bit like the wild west.”
If the SRAT rates the majority of detainees as high or extreme risk, it makes the concerns of this group of people easier to dismiss, he argues. “It becomes a whole lot easier to run a narrative that these are all hardened criminals.”
A 2020 report from the commonwealth ombudsman shows how easily someone can be rated as high risk. It detailed how detainees with any violent criminal history were assessed as high risk, no matter how much time had passed since the offence or any rehabilitation.
It also suggested that risk assessments could be escalated in “unwarranted” ways: someone convicted of having a traffickable amount of a drug is automatically linked to organised crime, which automatically raises the risk of violence. “Therefore, a person may be assessed as having links to organised crime and an associated higher risk rating without any material facts to support that rating,” it concluded.
Sarah, the former Serco analyst, told Guardian Australia that after the 2015 unrest on Christmas Island after the death of an Iranian detainee, a new directive was issued that anyone who had criminal history involving any violence was automatically rated high risk.
“It didn’t matter when the event occurred, it didn’t matter if they had served their time,” she said. There was little room for nuance: someone with a common assault charge and someone with a very long criminal history of violence would both be rated high. “It was a zero tolerance policy.”
Likewise, she said, anyone brought to a mainland detention centre from offshore was automatically assessed as high risk due to a perceived incentive for escape.
default
No process for review
When Jonathan Hall Spence, a lawyer with the Public Interest Advocacy Centre, obtained his client’s SRAT under FOI, it was heavily redacted. He describes the assessment tool as “an extremely opaque process”.
Hall Spence recently obtained a settlement from the government and Serco for his client, an asylum seeker named Yasir who arrived in Australia in 2013, over the use of handcuffs in detention. Yasir was forced to wear the restraints to attend essential medical appointments, even though he had been diagnosed with PTSD due to being tortured in his home country.
The terms of the settlement are confidential, but Hall Spence said the SRAT seemed to be producing “very conservative assessments of risk” that are unable to be analysed or challenged. “There’s no process for having it reviewed or overturned or reconsidered,” he said.
He’s also concerned about how static the tool is. “If you’re categorised as a person with high risk or extreme risk, you’re likely to unable ever be able to change that risk grading, even if you have very long periods of perfect behaviour,” he said. “We have serious doubts about the accuracy of the [SRAT].”
While someone’s risk assessment is meant to be reviewed monthly, it is exceedingly rare for it to be lowered, several people familiar with the system told Guardian Australia.
Sarah thinks SRATs could be an effective tool if analysts were able to use them correctly, but that became impossible. “The integrity of the tool was fractured, just through policy overrides,” she said, like the decision to rate anyone who had come from offshore detention as high risk. “It did become an abusive tool.”
Despite his victory with the information commissioner, Anees remains outraged by the control the SRAT had over his life, all without his knowledge.
“You don’t get justice, fair justice, where you get a say in what is written against your name.”
In a statement provided after publication, a Home Affairs spokesperson said: “The SRAT considers each detainee’s individual circumstances, including consideration of an individual’s capability (e.g. age, frailty, medical condition) and intent (e.g. immigration pathway, behaviour, prevalence of incidents), and is reviewed at regular intervals.”
Serco did not respond to a request for comment by deadline.
* Names have been changed.
Do you know more? Email abogle@protonmail.com
