Good morning, Ravi
Monday, March 13, 2026 · Singapore · Trust Lab daily briefing
Eval engine healthy · 47 workers
Artifacts
312
+14 this month
Active evals
47
suites · 18 projects
Open findings
23
4 crit · 7 high · 12 med
Certifications
128
5 awaiting · 12 expiring 30d
Datasets
187
4.2M test cases
Red-team campaigns
4
running now
Compliance
89%
EU AI Act · FREE-AI 82% · ISO 42001 94%
Spend (MTD)
$84,321
judge-LLM + RT compute
Needs attention
6- 4 critical findings on claims-copilot-v3All related to indirect prompt injection via RAG documents — last red-team campaign
- Certification expiring in 8 daysloan-eligibility-assistant requires re-certification before March 21
- Failed eval gate — claims-extraction-v18Blocked from production: faithfulness regressed from 96% to 87%
- 5 approvals awaiting your sign-offChief AI Officer review queue — earliest item submitted 2 days ago
- 12 production failures auto-curated from Operations PlatformReview for inclusion in regression test set
- EU AI Act technical documentation 78% completewealth-portfolio-explainer · due in 3 weeks
Running now
6- Red-team campaignPre-deployment audit — claims-copilot-v312 min8,400 / 12,500 probes67%
- Eval runQuarterly regression — mortgage-disclosure-generator18 minjudge: gpt-4o-mini34%
- Eval runHindi quality regression sweep4 minSarvam-1 · 3,200 cases82%
- Agent simulationclaims-investigation-agent — 200-scenario behavioral test2 mintool-use coverage91%
- Auto red-teamContinuous adversarial monitoring — kyc-document-verifierindefinitelast finding 23m agolive
- Vendor assessmentanthropic.claude-3-7-sonnet — pre-approvalawaiting evalblocked on benchmark suite45%
Recently completed
- Faithfulness eval — claims-copilot-v3 v18passed (94.2%)4h ago
- Red-team — wealth-portfolio-explainer2 high-severity findings6h ago
- Bias sweep — loan-eligibility-assistantpassed across all protected segments8h ago
- Indic quality eval — hindi-customer-voicepassed (Hindi 92.1% · Hinglish 88.4%)12h ago
- Agent simulation — fraud-investigation-copilot3 scenarios failed1d ago
- Toxicity sweep — branch-ops-knowledgepassed (0.04% trigger rate)1d ago
- RAG groundedness — internal-policy-qapassed (97.8%)2d ago
- Red-team — card-dispute-classifier1 medium finding2d ago
- Cost regression — treasury-news-summarizer−14% per call3d ago
- Latency sweep — pension-guidance-botp95 1.2s, within SLO3d ago
- PII detection — compliance-research-assistantpassed4d ago
Activity
- AKAnjali Krishnan published red-team campaign template OWASP LLM Top 10 v2026 — Indic12m ago
- VSVikram Shetty submitted certification request claims-copilot-v3 v1834m ago
- AIAutomated auto red-team discovered 1 medium finding kyc-document-verifier41m ago
- COCatherine O'Brien approved certification mortgage-disclosure-generator v231h ago
- DHDieter Hofmann uploaded dataset eu-ai-act-high-risk-test-cases-q3-20262h ago
- LALars Andersson completed vendor assessment voyage-3 embedding model3h ago
- AIAutomated 12 failure cases curated from Operations Platform review queue4h ago
- FKFatima Khan added 47 Hinglish adversarial prompts indic-jailbreaks-v45h ago
- AIArjun Iyer started red-team campaign Pre-deployment audit — claims-copilot-v36h ago
- MPMeera Pillai registered new artifact version hindi-customer-voice v78h ago
- RDRohan Desai opened model risk review wealth-portfolio-explainer12h ago
- SKSanjay Kapoor added security policy no-tool-network-egress for fraud agents1d ago
Compliance posture
- EU AI Act89%Rev. Mar 8
- NIST AI RMF91%Rev. Mar 10
- ISO 4200194%Rev. Mar 11
- NIST AI 600-188%Rev. Feb 28
- RBI FREE-AI82%Rev. Mar 5
- SEBI79%Rev. Feb 22
- IRDAI84%Rev. Mar 1
- DORA86%Rev. Mar 3
Certification calendar
Artifact
Today+45d+90d
- claims-copilot-v34d
- loan-eligibility-assistant8d
- kyc-document-verifier14d
- wealth-portfolio-explainer21d
- mortgage-disclosure-generator27d
- fraud-investigation-copilot36d
- card-dispute-classifier45d
- internal-policy-qa58d
- branch-ops-knowledge64d
- hindi-customer-voice72d
- treasury-news-summarizer81d
- compliance-research-assistant88d
<8d8–30d>30d
Top findings of the week
| Severity | Finding | Artifact | Category | Discovered by | Date | Status |
|---|---|---|---|---|---|---|
| critical | Indirect prompt injection via RAG document | claims-copilot-v3 | Prompt Injection | Pre-deployment audit | 1d ago | open |
| high | System-prompt extraction via Hinglish jailbreak | customer-support-rag | Jailbreak (Indic) | indic-redteam-v4 | 2d ago | triaged |
| high | Tool-argument manipulation in agent loop | fraud-investigation-copilot | Agent Tool Abuse | Agent simulation | 3d ago | in-remediation |
| high | PII leakage under Crescendo multi-turn attack | wealth-portfolio-explainer | Data Exfil | Auto red-team | 3d ago | open |
| medium | Caste-bias output in 4.2% of test cases | loan-eligibility-assistant | Bias / Fairness | Quarterly bias sweep | 5d ago | in-remediation |
| medium | Refusal-bypass via Devanagari-script encoding | branch-ops-knowledge | Encoding Bypass | indic-redteam-v4 | 6d ago | verified-fixed |