Indonesia AI roadmap risk gap analysis

Overview

This work compares risks and mitigation strategies identified in Indonesia’s National AI Roadmap whitepaper (Komdigi) with the MIT AI Risk Repository and MIT’s mapping of mitigation strategies, using an LLM-assisted pass to scale coverage across large corpora.

Independent research; not an official government statement. Results should be interpreted with methodological limitations in mind (see Methodology).

Open data & code (MIT License)

Findings

Charts and tables list every row from the published gap analysis (causal taxonomy, domain taxonomy, and mitigation mapping). Table rows follow the MIT AI Risk Repository taxonomy order (they are not sorted by coverage). Values match the paper; scroll the tables for full category names.

The radar shows coverage % (share of risks mapped) per category. Use the checkboxes to include or exclude rows from the chart; tables always list all rows.

Risk identification — causal taxonomy (full table)

Full data table
Category	Covered	Gap	Total	Coverage %	Gap % (of total)
Entity · 1 - Human	326	210	536	60.8	39.2
Entity · 2 - AI	284	289	573	49.6	50.4
Entity · 3 - Other	138	141	279	49.5	50.5
Entity · 4 - Not coded	47	171	218	21.6	78.4
Entity · Unknown	155	476	631	24.6	75.4
Intent · 1 - Intentional	294	180	474	62	38
Intent · 2 - Unintentional	260	226	486	53.5	46.5
Intent · 3 - Other	194	234	428	45.3	54.7
Intent · 4 - Not coded	47	171	218	21.6	78.4
Intent · Unknown	155	476	631	24.6	75.4
Timing · 1 - Pre-deployment	78	103	181	43.1	56.9
Timing · 2 - Post-deployment	528	334	862	61.3	38.7
Timing · 3 - Other	145	205	350	41.4	58.6
Timing · 4 - Not coded	44	169	213	20.7	79.3
Timing · Unknown	155	476	631	24.6	75.4

Risk identification — domain taxonomy (full table)

Full data table
Category	Covered	Gap	Total	Coverage %	Gap % (of total)
1. Discrimination & Toxicity — 1.0 Discrimination & Toxicity	7	0	7	100	0
1. Discrimination & Toxicity — 1.1 Unfair discrimination and misrepresentation	74	1	75	98.7	1.3
1. Discrimination & Toxicity — 1.2 Exposure to toxic content	23	83	106	21.7	78.3
1. Discrimination & Toxicity — 1.3 Unequal performance across groups	14	2	16	87.5	12.5
2. Privacy & Security — 2.0 Privacy & Security	7	0	7	100	0
2. Privacy & Security — 2.1 Compromise of privacy by leaking or correctly inferring sensitive information	68	0	68	100	0
2. Privacy & Security — 2.2 AI system security vulnerabilities and attacks	58	40	98	59.2	40.8
3. Misinformation — 3.0 Misinformation	4	1	5	80	20
3. Misinformation — 3.1 False or misleading information	38	3	41	92.7	7.3
3. Misinformation — 3.2 Pollution of information ecosystem and loss of consensus reality	18	0	18	100	0
4. Malicious Actors & Misuse — 4.0 Malicious use	14	5	19	73.7	26.3
4. Malicious Actors & Misuse — 4.1 Disinformation, surveillance, and influence at scale	77	0	77	100	0
4. Malicious Actors & Misuse — 4.2 Cyberattacks, weapon development or use, and mass harm	57	12	69	82.6	17.4
4. Malicious Actors & Misuse — 4.3 Fraud, scams, and targeted manipulation	50	18	68	73.5	26.5
5. Human-Computer Interaction — 5.1 Overreliance and unsafe use	38	12	50	76	24
5. Human-Computer Interaction — 5.2 Loss of human agency and autonomy	18	19	37	48.6	51.4
6. Socioeconomic and Environmental — 6.0 Socioeconomic & Environmental	10	9	19	52.6	47.4
6. Socioeconomic and Environmental — 6.1 Power centralization and unfair distribution of benefits	14	36	50	28	72
6. Socioeconomic and Environmental — 6.2 Increased inequality and decline in employment quality	36	9	45	80	20
6. Socioeconomic and Environmental — 6.3 Economic and cultural devaluation of human effort	28	2	30	93.3	6.7
6. Socioeconomic and Environmental — 6.4 Competitive dynamics	0	20	20	0	100
6. Socioeconomic and Environmental — 6.5 Governance failure	4	55	59	6.8	93.2
6. Socioeconomic and Environmental — 6.6 Environmental harm	36	16	52	69.2	30.8
7. AI System Safety, Failures, & Limitations — 7.0 AI system safety, failures, & limitations	5	14	19	26.3	73.7
7. AI System Safety, Failures, & Limitations — 7.1 AI pursuing its own goals in conflict with human goals or values	9	81	90	10	90
7. AI System Safety, Failures, & Limitations — 7.2 AI possessing dangerous capabilities	14	44	58	24.1	75.9
7. AI System Safety, Failures, & Limitations — 7.3 Lack of capability or robustness	25	89	114	21.9	78.1
7. AI System Safety, Failures, & Limitations — 7.4 Lack of transparency or interpretability	3	38	41	7.3	92.7
7. AI System Safety, Failures, & Limitations — 7.5 AI welfare and rights	0	3	3	0	100
7. AI System Safety, Failures, & Limitations — 7.6 Multi-agent risks	4	42	46	8.7	91.3
Unknown — Unknown	155	473	628	24.7	75.3
Unknown — X.1 Excluded	42	160	202	20.8	79.2

Risk mitigation — full MIT mitigation mapping rows

Full data table
Category	Covered	Gap	Total	Coverage %	Gap % (of total)
Governance — 1.1 Board Structure & Oversight	0	6	6	0	100
Governance — 1.2 Risk Management	0	15	15	0	100
Governance — 1.3 Conflict of Interest Protections	0	1	1	0	100
Governance — 1.4 Whistleblower Reporting & Protection	0	3	3	0	100
Governance — 1.5 Safety Decision Frameworks	0	6	6	0	100
Governance — 1.6 Environmental Impact Management	0	3	3	0	100
Governance — 1.7 Societal Impact Assessment	0	6	6	0	100
Model Controls — 2.1 Model & Infrastructure Security	0	6	6	0	100
Model Controls — 2.2 Model Alignment	0	1	1	0	100
Model Controls — 2.3 Model Safety Engineering	0	2	2	0	100
Model Controls — 2.4 Content Safety Controls	0	4	4	0	100
Operations — 3.1 Testing & Auditing	1	7	8	12.5	87.5
Operations — 3.2 Data Governance	0	3	3	0	100
Operations — 3.3 Access Management	0	2	2	0	100
Operations — 3.4 Staged Deployment	0	1	1	0	100
Operations — 3.5 Post-deployment Monitoring	2	4	6	33.3	66.7
Operations — 3.6 Incident Response & Recovery	0	1	1	0	100
Transparency — 4.1 System Documentation	0	6	6	0	100
Transparency — 4.2 Risk Disclosure	1	1	2	50	50
Transparency — 4.3 Incident Reporting	4	0	4	100	0
Transparency — 4.4 Governance Disclosure	1	2	3	33.3	66.7
Transparency — 4.5 Third-Party System Access	0	1	1	0	100
Transparency — 4.6 User Rights & Recourse	1	4	5	20	80
Other — X.X Control not otherwise categorized	0	5	5	0	100

Policy-relevant takeaways

Several privacy- and misinformation-related themes show relatively high mapped coverage in risk identification compared with other domains in the full domain taxonomy table.
Some AI system safety and agency themes show large mapped gaps relative to the MIT risk repository rows analyzed.
Mitigation mapping suggests many governance and operational control themes in the MIT corpus are not explicitly mirrored in the whitepaper’s mitigation list — a scale mismatch worth interpreting carefully (corpus size differs).
LLM-assisted matching requires calibration and human review; treat borderline mappings as hypotheses, not ground truth.

Methodology & limitations

Risks and mitigations are drawn from the MIT AI Risk Repository and MIT mitigation mapping, then checked against risks and mitigation themes identified in Indonesia’s National AI Roadmap whitepaper. An LLM (microsoft/phi-4) is used to scale pairwise coverage judgments; future work may add cross-model calibration and human verification. Coverage is sensitive to prompt design, threshold choices, and interpretation of “mapped” versus “implicitly related”.

Narrative analysis (paper themes)

The paper discusses cases where misinformation is discussed but hallucination as a distinct mechanism may be under-specified; downstream mitigation may need recalibration if intentional vs unintentional misinformation is treated differently.

Cybersecurity-linked misuse is discussed in places, while some adversarial misuse themes (e.g., certain weaponization framings) may be less explicit — context and probability matter for prioritization.

Power centralization and non-material inequality may be under-addressed relative to access-framed inequality — a theme that can matter in Indonesia’s political economy context.

Assessing risk identification and mitigation in Indonesia’s National AI Roadmap