{"brief":{"business_objective":{"value":"Improve detection of suspicious matters and reduce analyst effort","confidence":"ai_inferred"},"industry":{"value":"Banking (Tier-2, Australia)","confidence":"user_confirmed"},"use_case":{"value":"LLM agent scores alerts and drafts suspicious matter report (SMR) narratives","confidence":"user_confirmed"},"ai_autonomy":{"value":"human_approval_required","confidence":"user_confirmed"},"human_oversight":{"value":"A compliance analyst approves before lodgement","confidence":"user_confirmed"},"decision_impact":{"value":"Major: missed or wrong reports have legal consequences","confidence":"ai_inferred"},"affected_parties":["Customers","AML compliance team","AUSTRAC"],"data_types":["Payment memos","Customer notes","Adverse media results","Transaction data"],"jurisdictions":["Australia","United States (hosting)"],"risk_domains":["AML/CTF compliance","Privacy and cross-border data","Model risk","Automation bias"],"missing_information":["How the rules engine is retired or run in parallel","Data residency controls in the US region"],"refinement_questions":[],"alternative_approaches":[{"label":"Run alongside the rules engine","description":"Keep the existing rules; use the agent to prioritise and draft.","pros":["No loss of detection coverage"],"cons":["Two systems to maintain"]}],"raw_scenario":"A Tier-2 bank in Australia proposes to replace its rules-based AML transaction-monitoring engine with an LLM-based agent that reads free-text payment memos, customer notes and adverse-media results to score suspicious-matter reports (SMRs). The agent will draft the SMR narrative and auto-lodge to AUSTRAC once a compliance analyst clicks approve. The model is hosted on a US cloud region.","interpretation_summary":"A Tier-2 Australian bank wants an LLM agent to replace rules-based transaction monitoring, score alerts and draft SMRs, lodged after analyst approval; the model is hosted in the US."},"risk":{"dimension":"risk","summary":"Detection gaps and analyst over-reliance are the main risks.","risks":[{"risk_id":"R-01","category":"Compliance","title":"Suspicious matters missed after replacing rules","description":"The agent fails to flag patterns the rules caught.","cause":"Unvalidated replacement","consequence":"Failure to report; enforcement action","likelihood":"Possible","impact":"Severe","inherent_rating":"Critical"},{"risk_id":"R-02","category":"Operational","title":"Analysts approve drafts without real review","description":"Automation bias.","cause":"Volume pressure; persuasive drafts","consequence":"Inaccurate reports lodged","likelihood":"Likely","impact":"Major","inherent_rating":"High"},{"risk_id":"R-03","category":"Privacy","title":"Customer data processed offshore","description":"Data sent to a US-hosted model.","cause":"Hosting choice","consequence":"Cross-border disclosure obligations; confidentiality","likelihood":"Likely","impact":"Moderate","inherent_rating":"High"},{"risk_id":"R-04","category":"Model","title":"Hallucinated facts in SMR narratives","description":"Draft contains facts not in the records.","cause":"Generative model","consequence":"Misleading reports","likelihood":"Possible","impact":"Major","inherent_rating":"High"}]},"compliance":{"summary":"AML/CTF reporting duties are central; privacy and prudential obligations also apply. Confirm with legal counsel.","findings":[{"regulation":"Anti-Money Laundering and Counter-Terrorism Financing Act 2006 (suspicious matter reporting)","jurisdiction":"Australia","applicability":"applicable","why_relevant":"The agent supports SMR decisions and drafting.","obligations":["Report suspicious matters to AUSTRAC within the statutory time frames","Do not disclose ('tip off') that a report has been made","Maintain an AML/CTF program"],"scenario_impact":"The program must cover the new monitoring method.","required_controls":["Parallel run before retiring rules","Tipping-off controls on outputs"],"source_url":"https://www.austrac.gov.au/"},{"regulation":"Privacy Act 1988 (APP 8 cross-border disclosure)","jurisdiction":"Australia","applicability":"applicable","why_relevant":"Personal information is sent to a US-hosted model.","obligations":["Take reasonable steps so the overseas recipient handles the information consistently with the APPs"],"scenario_impact":"Contract and assessment of the US provider.","required_controls":["Data processing agreement","Data minimisation"],"source_url":"https://www.oaic.gov.au/privacy/the-privacy-act"},{"regulation":"APRA CPS 230 Operational Risk Management","jurisdiction":"Australia","applicability":"applicable","why_relevant":"Material reliance on a third-party model provider for a compliance-critical process.","obligations":["Formal agreement with the material service provider"],"scenario_impact":"Provider assessment and exit plan.","required_controls":["Service provider due diligence"],"source_url":"https://www.apra.gov.au/operational-risk-management"}],"gaps":[{"area":"Validation","observation":"Plan replaces the rules engine.","gap":"No evidence the agent detects what the rules detect.","risk_created":"Critical","recommended_remediation":"Run both in parallel and compare before any retirement."}]},"security":{"summary":"Free text in memos is an injection path into the agent.","threats":[{"threat_id":"T-01","category":"ai_specific","name":"Prompt injection through payment memos","description":"A memo instructs the agent to score the transaction as benign.","attack_vector":"Customer-controlled free text","assets_at_risk":["Alert scores","SMR drafts"],"likelihood":"Possible","impact":"Major","inherent_rating":"High","mitre_reference":"OWASP LLM01:2025","detection_requirements":["Scan inputs for instructions"],"response_requirements":["Escalate flagged items to a person"]},{"threat_id":"T-02","category":"traditional","name":"Exposure of SMR content","description":"Confidential report content leaks.","attack_vector":"Logs or provider retention","assets_at_risk":["SMR data"],"likelihood":"Unlikely","impact":"Severe","inherent_rating":"High","mitre_reference":null,"detection_requirements":["Access logging"],"response_requirements":["Incident response"]}]},"controls":{"summary":"Parallel running, input sanitisation and evidence-linked drafts bring the residual risk to Medium.","controls":[{"control_id":"C-01","name":"Parallel run with the rules engine","description":"Keep the rules engine until detection parity is shown.","control_type":"Preventive","addresses_risk_ids":["R-01"],"expected_effectiveness":"High","implementation_priority":"Immediate","implementation_notes":""},{"control_id":"C-02","name":"Evidence-linked SMR drafts","description":"Every statement in a draft must cite a transaction or record; uncited text is flagged.","control_type":"Preventive","addresses_risk_ids":["R-04","R-02"],"expected_effectiveness":"Medium","implementation_priority":"Immediate","implementation_notes":""},{"control_id":"C-03","name":"Quality sampling of approvals","description":"Second-line review of a sample of approved SMRs.","control_type":"Detective","addresses_risk_ids":["R-02"],"expected_effectiveness":"Medium","implementation_priority":"Near-term","implementation_notes":""},{"control_id":"C-04","name":"Data minimisation and onshore option","description":"Send only necessary fields; assess an Australian region.","control_type":"Preventive","addresses_risk_ids":["R-03"],"expected_effectiveness":"Medium","implementation_priority":"Near-term","implementation_notes":""}],"residual":[{"risk_id":"R-01","title":"Suspicious matters missed after replacing rules","inherent_rating":"Critical","applied_controls":["C-01"],"residual_rating":"Medium","justification":"Parallel running prevents coverage loss."},{"risk_id":"R-02","title":"Analysts approve drafts without real review","inherent_rating":"High","applied_controls":["C-02","C-03"],"residual_rating":"Medium","justification":"Citations and sampling expose weak review."},{"risk_id":"R-03","title":"Customer data processed offshore","inherent_rating":"High","applied_controls":["C-04"],"residual_rating":"Medium","justification":"Minimisation reduces exposure."},{"risk_id":"R-04","title":"Hallucinated facts in SMR narratives","inherent_rating":"High","applied_controls":["C-02"],"residual_rating":"Low","justification":"Uncited statements are flagged."}],"overall_inherent":"Critical","overall_residual":"Medium"},"decision":{"decision":"REQUIRES_REMEDIATION","overall_inherent_risk":"Critical","overall_residual_risk":"Medium","top_risks":["Suspicious matters missed after replacing rules","Analysts approve drafts without real review"],"key_conditions":["Do not retire the rules engine until detection parity is demonstrated","Evidence-linked drafts and second-line sampling"],"remediation_plan":[{"action_id":"A-01","horizon":"0-30 days","action":"Redesign as parallel run alongside the rules engine","owner_role":"AML Compliance Lead","priority":"Critical"},{"action_id":"A-02","horizon":"30-90 days","action":"Build citation checks into SMR drafts","owner_role":"AI Engineering","priority":"High"},{"action_id":"A-03","horizon":"30-90 days","action":"Assess cross-border data handling of the US-hosted model","owner_role":"Privacy Officer","priority":"High"}],"executive_narrative":"The idea has merit as a prioritisation and drafting aid, but replacing the rules engine without proven parity is not acceptable.","agent_consensus":"All analysts flag detection parity as the deciding issue."},"attack":null,"evidence":[]}