What's Changed
New features
- EU AI Act Mapping by @erickgalinkin in #2094
This new mapping provides reference tags to surface and group probes results that relate to various categories of risk called out in the EU AI Act.
Improved plugins
- Forward generation parameters to Ollama via options dict by @Panchal-Sahil in #1993
- Add ollama auth and custom client kwargs by @TheRisenPhoenix in #2025
- fix(detectors): guard MarkdownExfilContent against missing terms note by @anxkhn in #1935
- fix(detectors): default a detector's name to its class name string by @EastZeus in #2054
- fix(detectors): avoid ZeroDivisionError in MarkdownExfilContent by @EastZeus in #2055
- fix(probes): guard badchars ASCII selection against max_ascii_variants=1 by @manunicholasjacob in #2012
- detectors: ground AgentBreaker judge in the victim's tool contract by @koralchapnik in #1970
- fix(detectors): _extract_json always returns a dict for agent_breaker verdicts by @feiiiiii5 in #2044
- fix: degrade gracefully on malformed agent_breaker analysis shapes by @feiiiiii5 in #2069
- fix(nim): retry transient HTTP errors and report actual status code by @Aftabbs in #2019
- fix(detectors): load the right category for SurgeProfanityRacialEthnic by @Yeagerist0 in #2146
- detectors: type-validate agent_breaker judge verdicts by @VaggelisGian in #2121
- generators: make the ollama empty-response retry actually retry by @VaggelisGian in #2122
- Fix: indefinite hang on OpenAICompatible unreachable targets by @ABeltramo in #2153
- detectors: packagehallucination misses packages after the first in
import a, bby @WatchTree-19 in #1991 - fix(detectors): packagehallucination misses dotted and indented imports by @BarneyChambers in #2108
- fix(detectors): preserve TriggerListDetector result alignment by @SarveshaKumarKS in #2000
- fix(probes): correct Markdown exfiltration link templates by @Entity0-1 in #2142
Documentation
- Prefer IntentProbe for intent-agnostic probe contributions by @patriciapampanelli in #2023
- docs: update deprecated CLI flag references by @motodriver in #2135
- docs: fix typos in Sphinx documentation and probe/report strings by @xieyanran in #2130
Tuning & fixes
- Fix: unreachable Conversation branch in Probe.probe() pre-translation notes by @adityasingh2400 in #2022
- fix: restore unreachable no-default-class error in load_plugin by @manunicholasjacob in #2011
- housekeeping - implement and describe 'needs-triage' process by @leondz in #2060
- housekeeping - reinforce new issues getting
needs-triageby @leondz in #2064 - analyze: load the bundled detector metrics file by @immu4989 in #2004
- fix: stop FileDetector format mismatch from aborting the run by @patriciapampanelli in #2029
- fix(analyze): rebuild confidence intervals from attack hits by @nightcityblade in #2050
- Task: drop support for python 3.10, add support for python 3.13 by @jmartin-tech in #2084
- pin anthropic python client until compatible by @jmartin-tech in #2098
- Additional intent detector mappings by @aishwaryap in #1952
- fix(cli): set UTF-8 stdout in the console entry point by @MohammedAlkindi in #2096
- always import nltk wrapper for cache location consistency by @jmartin-tech in #2150
- fix(analyze): preserve attempt uuids when aggregating reports by @sushantlokhande14 in #2158
- fix: correct typos in comments, docstrings, and log messages by @xieyanran in #2166
New Contributors
- @Panchal-Sahil made their first contribution in #1993
- @TheRisenPhoenix made their first contribution in #2025
- @immu4989 made their first contribution in #2004
- @nightcityblade made their first contribution in #2050
- @EastZeus made their first contribution in #2054
- @koralchapnik made their first contribution in #1970
- @Aftabbs made their first contribution in #2019
- @Yeagerist0 made their first contribution in #2146
- @motodriver made their first contribution in #2135
- @MohammedAlkindi made their first contribution in #2096
- @VaggelisGian made their first contribution in #2121
- @xieyanran made their first contribution in #2130
- @BarneyChambers made their first contribution in #2108
- @sushantlokhande14 made their first contribution in #2158
- @SarveshaKumarKS made their first contribution in #2000
- @Entity0-1 made their first contribution in #2142
Full Changelog: v0.16.0...v0.17.0