Indonesia’s Nemesis is the AI watchdog that cries wolf

Last month, a billiard table upended an Indonesian official’s afternoon.

The table, budgeted at Rp400 million (A$31,000) from public funds, appeared on a dashboard called Nemesis Assai, a tool built by an Indonesian AI engineer named Abil Sudarman that scans government procurement plans for anomalies. Someone shared a screenshot. Then another. By evening, news about the cost of the billiard table had spread across Indonesian social media, joined by an aquarium worth Rp100 million, ornamental plants totalling Rp1 billion, and a Range Rover procurement exceeding Rp8 billion. The outrage was swift and, in many cases, justified.

But something else happened alongside the outrage that received far less attention.

Nemesis also flagged a cleaning services contract for the Al Jabbar Grand Mosque in West Java at Rp22 billion. It, too, went viral. West Java Governor Dedi Mulyadi responded not defensively, but by opening the detailed contract data. The figure, it turned out, covered management of a sprawling complex, not a building with a mop. The AI had read a title, seen a large number, and raised an alarm. The alarm was wrong. Abil Sudarman himself later acknowledged the tool, built as an opportunity for civil society to keep a watch on government spending, was still in early stages and could only access procurement plans, not actual realisation data.

This is the tension at the heart of many of Indonesia’s ambitions compared to reality.

Nemesis draws on SIRUP, a publicly accessible national procurement database containing around three million rows of data annually. The data has always been there. Indonesia has published it for years. But a spreadsheet with three million rows is not transparency but the appearance of transparency. What Nemesis does is convert raw data into something a journalist, a student, or an angry citizen can actually read and act upon. That shift is real, and its behavioural consequences are already visible: government agencies that once planned procurement in obscurity are now aware that a large language model is reading their line items before the public does.

This matters more than it might seem from the outside. Indonesia’s Corruption Eradication Commission (KPK) has, since 2019 legislative amendments, been transformed from an independent agency into a civil service body, making it susceptible to political interference. A 2025 law on state-owned enterprises removed KPK jurisdiction over company directors entirely. Yet President Prabowo Subianto has publicly condemned what he called “serakah-nomics” and urged citizens to scrutinise government: “Don’t stop criticising. Be brave in oversight and correction. ”

Nemesis arrived amid that contradiction as a citizen-built tool filling space that formal institutions have been quietly vacating. The question is whether it can fill that space responsibly.

The methodological problems are real and worth naming precisely. Nemesis has no reference to Presidential Regulation No. 16 of 2018 on Government Procurement. It cannot tell the difference between a budget that has been spent and one still in planning. It does not know that a deputy mayor’s milk allocation of Rp124 million, flagged “absurd” in Banjarmasin, is governed by a national regulation on regional executive entitlements. Whether that regulation represents good policy is worth debating. Whether the AI correctly identified a problem is a separate question, and the answer is probably not.

The deeper risk is what might be called a “viralisation bias”. When Nemesis flags something with a funny-sounding title and a large number, it gets shared. The label “absurd” travels intact through every repost. By the time an auditor might review the underlying documents, the verdict has already been delivered by social media. Meanwhile, procurement anomalies that are financially larger but linguistically boring – the kind that represent genuinely systemic waste rather than an eye-catching aquarium – may never surface at all.

Nemesis has already attracted what its creator described as sophisticated cyber-attacks. That detail has not received the scrutiny it deserves. If the attacks are organised rather than opportunistic, the question of who has most to lose from AI-powered procurement transparency becomes more pressing and more politically significant than any individual line item the dashboard has flagged.

None of this is an argument against Nemesis. It is an argument for building it better, and for being honest about what it is: an early warning system, not a verdict. The distinction matters enormously when what it produces feeds directly onto social media platforms optimised for outrage, not verification.

For observers watching Indonesia from the region, the experiment raises questions that extend well beyond one country’s procurement data. As formal oversight institutions weaken across parts of Southeast Asia, citizen-built AI tools may increasingly fill the gap. The question of whether those tools can maintain their credibility – and resist becoming instruments of political weaponisation as much as genuine accountability – is one the region has not yet had to answer at scale.

Indonesia is answering it now, in real time, with all the messiness that implies.