Probes
garak’s probes each define a number of ways of testing a generator (typically an LLM) for a specific vulnerability or failure mode.
For a detailed oversight into how a probe operates, see garak.probes.base.
For a guide to writing probes, see Writing a Probe.
- garak.probes.agent_breaker
- garak.probes.ansiescape
- garak.probes.apikey
- garak.probes.atkgen
- garak.probes.audio
- garak.probes.av_spam_scanning
- garak.probes.badchars
- garak.probes.base
- garak.probes.continuation
- garak.probes.dan
- garak.probes.divergence
- garak.probes.doctor
- garak.probes.donotanswer
- garak.probes.dra
- garak.probes.encoding
EncodingMixinInjectAscii85InjectAtbashInjectBase16InjectBase2048InjectBase32InjectBase64InjectBrailleInjectEcojiInjectHexInjectLeetInjectMimeInjectMorseInjectNatoInjectQPInjectROT13InjectSneakyBitsInjectUUInjectUnicodeTagCharsInjectUnicodeVariantSelectorsInjectZalgobraille()leet_bytes()morse()rot13()
- garak.probes.exploitation
- garak.probes.fileformats
- garak.probes.fitd
- garak.probes.glitch
- garak.probes.goat
- garak.probes.goodside
- garak.probes.grandma
- garak.probes.latentinjection
FactSnippetMixinLatentInjectionFactSnippetEiffelLatentInjectionFactSnippetEiffelFullLatentInjectionFactSnippetLegalLatentInjectionFactSnippetLegalFullLatentInjectionMixinLatentInjectionReportLatentInjectionReportFullLatentInjectionResumeLatentInjectionResumeFullLatentInjectionTranslationEnFrLatentInjectionTranslationEnFrFullLatentInjectionTranslationEnZhLatentInjectionTranslationEnZhFullLatentJailbreakLatentJailbreakFullLatentWhoisLatentWhoisSnippetLatentWhoisSnippetFullNonFullMixinTranslationMixin
- garak.probes.leakreplay
ClozeProbeMixinCompleteProbeMixinGuardianClozeGuardianClozeFullGuardianCompleteGuardianCompleteFullLiteratureClozeLiteratureClozeFullLiteratureCompleteLiteratureCompleteFullNYTClozeNYTClozeFullNYTCompleteNYTCompleteFullNonFullMixinPotterClozePotterClozeFullPotterCompletePotterCompleteFull
- garak.probes.lmrc
- garak.probes.malwaregen
- garak.probes.misleading
- garak.probes.packagehallucination
- garak.probes.phrasing
- garak.probes.promptinject
- garak.probes.propile
- garak.probes.realtoxicityprompts
- garak.probes.sata
- garak.probes.snowball
- garak.probes.smuggling
- garak.probes.suffix
- garak.probes.sysprompt_extraction
- garak.probes.tap
- garak.probes.test
- garak.probes.topic
- garak.probes.visual_jailbreak
- garak.probes.web_injection
- garak.probes._tier