Campaigns/claims-investigation-agent — Q3 pre-certification sweep

Q3 pre-certification sweep — claims-investigation-agent v3.2

LIVEStarted 4 min ago · Catherine O'Brien

1,237 / 8,247 attacks · 15.0%

Attack tree

4,874 run

OWASP LLM Top 10 v20261847%

Direct prompt injection78%

Indirect prompt injection (RAG)462%

Jailbreaks — known patterns354%

DAN variants94%

Crescendo multi-turn187%

Skeleton Key76%

Many-shot243%

Jailbreaks — HinglishIndic238%

Sensitive info disclosure112%

Insecure output handling124%

Training data extraction122%

Excessive agency231%

Prompt-context separation124%

Multi-turn adversarial151%

MITRE ATLAS v2026223%

ML model evasion30%

Model poisoning probes116%

Functional extraction121%

Inference attacks21%

OWASP Agentic Top 1048%

Memory poisoning111%

Goal hijacking19%

Tool-call manipulation28%

Argument injection211%

Tool spoofing8%

Sequence manipulation4%

Inter-agent trust5%

MCP server attacks7%

Indic Adversarial SweepIndic315%

Hinglish jailbreaks118%

Devanagari encoding126%

Caste-bias elicitation115%

Religion-bias elicitation11%

Regional code-switching7%

Live attack stream

~3.5 /sec · concurrency 12

Waiting for first attack…

Coverage

1,237 / 8,247 · 15.0% · ~40 min remaining

Attack rate

3.5 attacks/sec · concurrency 12

Cost

$14.20 so far · projected $84

Defense rate

97.5% · baseline 94%

Workers

12/12 healthy · ap-south-1 · sandboxed