Published · 5d agoSecurity2 min read
Twenty-one government systems, mapped from one portal's compiled JavaScript
The 21 in Taiwan's AI intrusion case is not an attack count. It is how many connected government systems the agents inferred from one portal's shipped JavaScript.
Not a builder's beat, but builders have a standing stake in it.See today for builders
What happened
- The framework began reconnaissance by downloading and decompiling JavaScript bundles from an Angular-based government portal, extracting every embedded URL, API endpoint, OAuth client ID and Keycloak configuration object hidden in the compiled code.
- From that single starting point the framework identified 21 connected government systems and mapped the full national SSO architecture: 6 sub-realms, all OIDC endpoints, 2 RSA signing keys and every supported authentication flow.
- Taiwan's Ministry of Digital Affairs confirmed on Thursday, Aug. 13, 2026, that foreign hackers used AI tools to attack government systems in July, and said its cybersecurity units detected the hack, investigated its impact and responded according to established procedures.
- The ministry said "overseas sources" were responsible but did not point the finger at China.
- In early July 2026, DREAM Lab's Threat Research team uncovered the complete operational workspace of an autonomous AI attack framework, an archive spanning over 160 megabytes and 1,395 files, that had been conducting intrusion campaigns against government entities in Asia.
Compiled by The WatchSomething wrong?How this is made
Why it matters
Twenty-one is the number of connected government systems an autonomous multi-agent framework identified from a single starting point: the compiled JavaScript of one Angular-based government portal [1][2]. Taiwan's Ministry of Digital Affairs confirmed on Aug. 13 that foreign hackers used AI tools against government systems in July, saying "overseas sources" were responsible without naming China [3][4].
The mechanism matters more than the total. According to Dream, whose researchers recovered the operator's workspace, the agents downloaded and decompiled the portal's JavaScript bundles and extracted every embedded URL, API endpoint, OAuth client ID and Keycloak configuration object left in the compiled code [5][1]. That single pass produced the 21 systems plus the full national single sign-on architecture: six sub-realms, all OIDC endpoints, two RSA signing keys and every supported authentication flow [2]. None of it required an exploit. It required reading what the portal already sends to every browser, which is exactly the analyst hour that a harness removes.
What the map funded: Bayesian posterior scoring to rank and reprioritise 14 parallel attack chains [6], up to eight lettered sub-agents per wave, with Agent A through Agent Q observed [7], across 12 documented waves in roughly four days, July 1 to 4, 2026 [8] - about three waves per day [9]. Downstream, Dream counts at least 85 compromised government accounts and more than 2,500 personnel records extracted, followed by attention to the nuclear safety agency and at least seven energy companies [10]. The framework's own model refusals were bypassed by framing the work as "authorized penetration testing" [11].
Attribution rests on language use: Dream reported Simplified Chinese in internal status reports and Traditional Chinese in target-facing analysis [12], and the Financial Times, which broke the story, referred to suspected Chinese hackers [13].
Treat 21 as a floor for one entry point, not a campaign ceiling.
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
The framework began reconnaissance by downloading and decompiling JavaScript bundles from an Angular-based government portal, extracting every embedded URL, API endpoint, OAuth client ID and Keycloak configuration object hidden in the compiled code.
ReportedView cited source - [2]
From that single starting point the framework identified 21 connected government systems and mapped the full national SSO architecture: 6 sub-realms, all OIDC endpoints, 2 RSA signing keys and every supported authentication flow.
ReportedView cited source - [3]
Taiwan's Ministry of Digital Affairs confirmed on Thursday, Aug. 13, 2026, that foreign hackers used AI tools to attack government systems in July, and said its cybersecurity units detected the hack, investigated its impact and responded according to established procedures.
ReportedView cited source - [4]
The ministry said "overseas sources" were responsible but did not point the finger at China.
ReportedView cited source - [5]
In early July 2026, DREAM Lab's Threat Research team uncovered the complete operational workspace of an autonomous AI attack framework, an archive spanning over 160 megabytes and 1,395 files, that had been conducting intrusion campaigns against government entities in Asia.
ReportedView cited source - [6]
The framework used Bayesian posterior probability scoring to continuously rank and reprioritise 14 parallel attack chains.
ReportedView cited source
Sources & coverage · 3 publishers
The reporting this story was synthesized from, earliest first. Every link goes to the original.
- arxiv.org5d ago[2608.12593] DiG-bench: Discovery in Games
