Legal Experts Warn Home Office AI Tools Risk Unlawful Asylum Assessments

(UNITED KINGDOM) — The UK Home Office used AI tools to assist in asylum assessments, and a legal opinion by two senior lawyers warned the practice is “likely to be unlawful” because applicants are not told the technology is being used and may not get a fair chance to correct mistakes.

Robin Allen KC and Dee Masters of Cloisters Chambers said the way the Home Office has deployed AI in asylum casework raises concerns about transparency, procedural fairness and data protection, in a process where credibility findings and factual summaries can shape life-changing decisions.

Their legal opinion said asylum applicants are not being informed that AI tools are being used to assess their cases, and are not being given opportunities to correct errors generated by those systems.

That, the opinion argued, is “likely to be unlawful” as a matter of procedural fairness, and could also breach data protection law.

The warning comes as the Home Office faces a backlog of asylum cases and looks for administrative ways to process claims more quickly, while lawyers stress that speed cannot come at the expense of fairness when decisions turn on consistency, detail and credibility.

Allen and Masters focused on what they described as missing safeguards around notice and accountability, including how applicants would ever learn that AI contributed to the information used in their assessment.

They also highlighted the practical difficulty of challenging adverse material if a claimant does not know how a summary was produced, what information it relied on, or where an error entered the file.

The legal opinion said the Home Office’s approach creates a risk that applicants cannot meaningfully respond to potentially harmful inaccuracies, a core element of procedural fairness in decision-making.

In parallel, the opinion flagged data protection concerns that can arise when automated tools handle personal information in ways that are not clearly explained to the person concerned.

Those concerns included transparency duties and accuracy obligations, as well as the need for safeguards when errors may affect outcomes.

The lawyers’ criticism centred on the use of AI to support core casework functions, including summarising information used in asylum assessments and producing decision-support material inside the system.

The Home Office’s own evaluation of its AI systems, cited in the legal opinion, found notable reliability problems during pilots and user testing.

In one pilot involving an AI summary tool, the evaluation found some AI-generated summaries were so flawed they had to be removed.

Feedback on another system also recorded that some users reported they were not confident in the tool’s accuracy.

Allen and Masters warned that errors in summaries can become “material” in asylum decision-making, because those documents can influence how caseworkers understand an account and how they test credibility.

“The legal opinion warns that ‘given the apparent inaccuracies in the summaries generated by the APS and ACS, there is a significant risk that decisions which are based on those summaries will be based upon and vitiated by material errors of fact,’” the opinion said.

Even when a human decision-maker remains formally responsible, the lawyers said, workflows can still create pressure to rely on machine-generated outputs that appear authoritative in internal notes.

They framed that risk as especially acute in asylum assessments, where credibility determinations often depend on careful attention to detail and the avoidance of misunderstandings that can harden into a negative view of an applicant’s account.

Allen and Masters said AI tools can have a place in administration if used properly, but said the current approach fails to meet the safeguards required when technology touches assessments that can affect an individual’s safety and future.

“‘Technology can assist decision-making, but it cannot undermine the careful human judgment required in asylum cases. Where AI tools are used without adequate safeguards, there is a real risk that unlawful or unfair decisions may result,’” they said.

They also called for fuller disclosure of how such systems function within the asylum process, including what role the outputs play and how caseworkers are expected to use them.

“‘If AI tools are influencing asylum decisions, there must be full transparency about how those systems operate and how their outputs are used,’” Allen and Masters said.

A central concern in the opinion was that people seeking asylum may not know AI tools are part of the assessment process at all.

Without that knowledge, the lawyers argued, applicants may not understand why a case file contains certain wording, summaries, or apparent inconsistencies, and may not be able to identify where a mistake has been introduced.

That matters, the opinion suggested, because an inaccurate summary can shape the framing of an interview, the interpretation of an answer, or the way an account is recorded in internal systems.

The legal criticism also went to the ability to challenge errors.

If an applicant is not told that AI was used, and is not shown what the system produced, the opinion said they may not have a realistic opportunity to correct the record before an adverse decision is made.

The lawyers presented that as a procedural fairness problem, because fairness includes a meaningful chance to understand and respond to adverse material that may affect a decision.

They also linked the issue to data protection questions, because personal information used in asylum assessments is sensitive, and the legal opinion warned of potential breaches if transparency and accuracy duties are not met.

The Home Office evaluation findings cited by the lawyers added weight to the concern, because they indicate errors are not hypothetical.

The pilots and user feedback described in the opinion pointed to quality failures serious enough that some outputs were removed, and to user doubts about whether the tools produced reliable results.

In a system that depends on accurate capture of an individual’s account, those findings can raise questions about whether AI summaries should be used at all, at least in their current form.

Allen and Masters argued that the combination of limited transparency and demonstrated error rates makes it harder to treat AI-generated material as a neutral administrative aid.

Instead, they said, inaccuracies can travel through a case file and influence later steps, including how evidence is weighed and how credibility is judged.

The opinion did not frame the issue as a narrow technical glitch, but as a governance and accountability problem in which the Home Office must decide what role, if any, AI tools should play in asylum assessments.

The lawyers called for “an immediate ban” on these tools, saying alternative methods exist to tackle the Home Office backlog while maintaining decision quality.

Their position, set out in the legal opinion, was that pausing or stopping the tools is necessary because the risk of unlawful or unfair decisions is too high without proper safeguards.

The opinion’s criticisms focus on safeguards that go beyond general assurances that humans remain in charge.

In practice, they said, the Home Office must ensure transparency about the existence and function of AI tools, and must set out how applicants can see, challenge and correct errors that might affect them.

They also called for accountability measures that show how AI outputs are used, including clear documentation standards and mechanisms for dealing with mistakes.

The legal opinion treated these issues as central to lawful decision-making, not optional additions.

For asylum applicants, the lawyers’ argument implies that knowing whether AI played a role is directly connected to the ability to defend the accuracy of their account.

For the Home Office, the opinion suggests that continuing to use AI without clear notice and correction routes may increase legal exposure, because decisions can be challenged on procedural fairness grounds if the process does not give a meaningful chance to respond to adverse material.

The concerns also go to data protection governance, because the opinion warned of potential breaches if the Home Office cannot show a lawful and transparent basis for how it processes and summarises information with AI tools.

The Home Office’s own evaluation findings, as described in the opinion, also suggest operational risks.

If caseworkers do not trust the outputs, or if some outputs are so flawed they must be removed, the tools may add work rather than reduce it, while still creating a risk that inaccurate content appears in internal files.

At the same time, the lawyers acknowledged that technology can assist decision-making, but drew a line at any use that erodes “the careful human judgment required in asylum cases.”

Their analysis placed weight on the sensitivity of asylum decisions, where the record of an applicant’s account, and the way it is summarised, can influence whether decision-makers view the account as consistent and credible.

Allen and Masters framed their call for transparency as a threshold requirement.

If AI tools influence asylum assessments in any way, they argued, applicants and the public must be able to understand how the tools operate and how the outputs are used, rather than treating the process as a closed internal workflow.

They also tied transparency to accountability, because it can enable oversight and allow individuals to identify and correct errors before those errors become embedded in the reasons for a decision.

The legal opinion’s conclusions now place attention on what steps the Home Office takes next, including whether it pauses the current use of AI tools or changes how it informs applicants and manages quality.

The opinion also points toward the type of safeguards that may become central in any future framework for AI in asylum assessments, including notice to applicants, explainability about system use, and practical correction mechanisms when errors arise.

Allen and Masters presented the issue as one of governance as well as technology, arguing that without safeguards, the Home Office risks decisions that are unlawful or unfair.

“‘Technology can assist decision-making, but it cannot undermine the careful human judgment required in asylum cases. Where AI tools are used without adequate safeguards, there is a real risk that unlawful or unfair decisions may result,’” they said.