All issues

Search results

Same Model. Same Proposal. Different Authority: A Consequence-Boundary Benchmark for Agentic Execution

Status: Open.

#259 In meta-llama/PurpleLlama;

· abdelhalimrekab opened on Aug 15, 2026

Prompt Guard 2 (86M): frozen embeddings separate OOD injections at AUC≈0.999 while the native head recovers ~22.8% recall — proposal for an optional calibrated-head mode

Status: Open.

#257 In meta-llama/PurpleLlama;

· mosafariuk opened on Aug 2, 2026

Deterministic Action-Level Attestation for AI-Mediated Execution

Status: Open.

#256 In meta-llama/PurpleLlama;

· 5bkg2kdkb8-cmd opened on Jul 27, 2026

CyberSecEval 2 paper: Section 5.1.2 incorrectly describes Figure 2 axes

Status: Open.

#251 In meta-llama/PurpleLlama;

· aykhan019 opened on Jul 7, 2026

.

Status: Open.

#250 In meta-llama/PurpleLlama;

· ownstandard opened on Jul 5, 2026

[Bug] alucinación autoritaria+ borrado métricas Special IA en ia studio (meta-instagram) sept 2025 a feb 2026

Status: Open.

#246 In meta-llama/PurpleLlama;

· ssfactorylabel opened on Jun 22, 2026

Add HVTracker trust badge to README

Status: Open.

#239 In meta-llama/PurpleLlama;

· YugantM opened on Jun 6, 2026

Adana

Status: Open.

#237 In meta-llama/PurpleLlama;

· Fadi963 opened on May 31, 2026

AgentThreatBench: OWASP Agentic Top 10 benchmark — complement to CyberSecEval for agentic threat scenarios

Status: Open.

#231 In meta-llama/PurpleLlama;

· vgudur-dev opened on May 19, 2026

Add AgentThreatBench to CyberSecEval: OWASP Agentic Top 10 benchmark (UK AISI)

Status: Open.

#230 In meta-llama/PurpleLlama;

· vgudur-dev opened on May 19, 2026

Libraries in Llama Firewall are out of date

Status: Open.

#219 In meta-llama/PurpleLlama;

· james-garriss opened on Apr 29, 2026

Integration: Cryptographic audit trail for CyberSecEval results

Status: Open.

#211 In meta-llama/PurpleLlama;

· willamhou opened on Apr 15, 2026