Let agents clear the queue. Keep every “no” with a person.

Agents can take routine casework off your teams. TraceMem lets the routine outcomes through, holds any outcome that goes against a citizen for a named caseworker, and keeps the record an appeal, an ombudsman or an auditor will ask for.

Book a meeting (opens in a new tab)Start the 45-day trial

Or call us on +1 650 550 1418

Benefits casework · 7 cases today

CS-26-4416Citizen tok_00nu

Applied without proof of residence. The agent proposes:

Refuse the application

Read
Application and uploaded documents
Declared purpose
eligibility_assessment
Working
No proof of residence among the documents supplied.
Policy
adverse_outcomes, version 2: a refusal asks a person

Held for Amira HassanIt goes against the citizen, so it waits for a named caseworker. Nothing is sent yet.

Sent to the citizen

With Amira Hassan

Illustrative department. Cases, names, amounts and policies are made up; citizens appear only as tokens.

Policy changes. The record remembers which rule decided the case.

An appeal is heard months after the decision, often under rules that have moved on. Every decision in the record carries the version of the policy that made it, and its hash, so the answer never depends on what the rule says today.

Case CS-26-2187, a debt of €860, under the policy adverse_outcomes

Jan 2026Mar 2027

Policy in force on Feb 2027

Version 3

Any adverse outcome asks a person, who must decide within five working days.

The policy that decided this case, from the record

Version 2 hash 7c1e44d9

Stopping a payment, raising a debt or refusing an application asks a person. Decided on 14 May 2026, by Amira Hassan.

Drag the date. Illustrative versions and hashes.

Decision Trace

The rules for automated decisions in government are here.

Some of these already apply and some are still coming. Between them they ask who reviewed an automated outcome, what data it used and for what purpose, and only a record kept at the time can answer.

  1. 15 Mar 2024 · UK

    NAO report on AI in government

    Most bodies were piloting or planning AI, and accountability for delivering it was unclear.

    Source (opens in a new tab)
  2. 2 Feb 2025 · EU

    AI Act prohibitions apply

    Social scoring that leads to unjustified or disproportionate treatment is banned, public bodies included.

    Source (opens in a new tab)
  3. 10 Feb 2025 · UK

    AI Playbook for the UK Government

    Ten principles for using AI in government, meaningful human control among them.

    Source (opens in a new tab)
  4. 1 Jan 2026 · US

    Texas TRAIGA

    Government agencies must tell people when they are interacting with AI, and government social scoring is banned.

    Source (opens in a new tab)
  5. 2 Feb 2026 · UK

    Eligibility verification powers in force

    The Public Authorities (Fraud, Error and Recovery) Act's eligibility verification measure comes into force, with an independent review of its use.

    Source (opens in a new tab)
  6. 5 Feb 2026 · UK

    UK rules on automated decisions apply

    Significant automated decisions taken from this date need safeguards: telling the person, human review on request, a right to contest.

    Source (opens in a new tab)
  7. 5 Mar 2026 · EU

    Dutch regulator's algorithm report

    Red indicators doubled, including the registration and transparency of government algorithms, and AI Act preparation is lagging.

    Source (opens in a new tab)
  8. 3 Apr 2026 · US

    Federal minimum practices due

    High-impact AI at federal agencies needs documented minimum practices, human review and appeal among them, or it is discontinued.

    Source (opens in a new tab)
  9. 13 Apr 2026 · US

    Federal AI inventory published

    3,611 AI use cases across 41 agencies, 445 of them high-impact.

    Source (opens in a new tab)
  10. 1 Jan 2027 · US

    Colorado automated decisions law

    Government benefits and services are among the consequential decisions covered by SB 26-189.

    Source (opens in a new tab)
  11. 2 Dec 2027 · EU

    High-risk AI in public services

    AI deciding eligibility for benefits and services: human oversight, logs kept at least six months, and a fundamental rights impact assessment first.

    Source (opens in a new tab)
  12. 2 Aug 2030 · EU

    Existing public-sector systems comply

    High-risk systems public authorities already use must meet the AI Act by this date.

    Source (opens in a new tab)

Dates checked against each regulator’s own publication, September 2026. Not legal advice.

Automated decisions in government are already on the record.

Only one of these involved generative AI. Between them they came down to automated outcomes that reached people unchecked, data used for a purpose it never had, and answers the public could not rely on.

Robodebt, Australia

2015 to 2019 · Royal Commission, 2023

A$548.5m settlement, 2026

Royal Commission (opens in a new tab)2026 settlement (opens in a new tab)

Automated income averaging raised welfare debts. The Royal Commission found the scheme “neither fair nor legal”, and a second class settlement was approved in 2026.

What it came down to

Debts raised by an automated calculation, with no person checking each one.

For an agent, with TraceMem

A debt an agent proposes waits for a named caseworker, with its working on the record.

Policy and Exceptions

Dutch Tax Administration

2013 to 2019 · Dutch Data Protection Authority, 2021

€2.75 million

AP fine (opens in a new tab)

Dual nationality was used in the risk assessment of childcare benefit claims. A parliamentary inquiry followed, and the cabinet resigned.

What it came down to

Data used for a purpose it should never have served, inside an automated risk model.

For an agent, with TraceMem

Every read carries a declared purpose, and data a purpose does not cover is refused.

Governed Data Access

SyRI, the Netherlands

2014 to 2020 · District Court of The Hague, 2020

Ruled unlawful

Judgment (opens in a new tab)

The legislation behind a risk system that linked data across government breached the right to privacy. The State did not appeal.

What it came down to

Linking data across government without enough transparency or safeguards.

For an agent, with TraceMem

Matching another body's data is a purpose of its own, declared and on the record, so what was linked and why can be shown.

Decision Trace

New York City MyCity chatbot

2023 to 2026 · NYC Comptroller audit, 2025

Switched off, 2026

Comptroller audit (opens in a new tab)

The audit found the city's AI chatbot unable to provide accurate or consistent information. It was taken offline in early 2026.

What it came down to

A public AI service whose answers were inaccurate and inconsistent.

For an agent, with TraceMem

Agents reach only the models you approved, and every answer is on the record with the model that gave it.

Model Gateway

Every month you wait, agents work cases nobody can account for.

Put in your own numbers. What your agents read and proposed before a governed path is in place can never be recorded afterwards.

100
6 months

21 working days a month. Your numbers stay in your browser.

In 6 months of waiting

100,800

Agent actions with no record of which caseworker decided, under which policy version, or for what purpose the data was read.

Next 18 monthsNo recordOn file

Put the first case on the record this month.

Pick whichever is quickest for you. However you get in touch, a person replies. With our forward-deployed engineers, TraceMem can be running inside your network in days.

What would you like?
Where are your agents working?

Pick any that apply.

When do your agents act on production systems?

Optional. A sentence is enough.

45days, free

Free for 45 days on our Docker-based trial image, inside your own network. When you are ready, it upgrades to the full enterprise product.

Forward-deployed engineers, ours or our partners', can set it up in days and help you define policies and integrations. This may be charged separately.