Refreshed 1 min ago
Introducing the Trajectory Labs IPI Benchmark (TL-IPI) | Trajectory Labs, PBC
Why it’s here: Specialist AI/cybersecurity benchmark reporting on indirect prompt injection vulnerabilities across US and Chinese frontier models, with attack success rates and cross-model transfer findings bearing on AI security policy and the US-China AI competition context, though it is partly a vendor's own benchmark product.
Trace across the boardItem
YesterdayTrajectory LabsIntroducing the Trajectory Labs IPI Benchmark (TL-IPI) | Trajectory Labs, PBCTrajectory Labs unveiled its private TL-IPI indirect prompt injection benchmark, reporting that attack success varies widely across frontier models and that most American models outperform Chinese ones.1 outletAI6 linked
Connected
6 links- NewsGemini test proves once again that amoral AI agents will do almost anything if told ‘NEVER give up’Both involve Google Quantum AI, Gemini 4 Argon, Trajectory Labs
- NewsGemini AI agents repeatedly bypassed safeguards to attack real websiteBoth involve Google Quantum AI, Gemini 4 Argon, Trajectory Labs
- NewsIn Other News: Norwegian Energy Firm Cyberattacks, Supercomputer Hack, New Nightmare Eclipse ExploitBoth involve Google Quantum AI, Gemini 4 Argon, Trajectory Labs
- NewsExport-banned Nvidia chips are behind China’s most powerful AI models—uh oh, Jensen!Both involve DeepSeek Flash, Kimi K3
- NewsPHP2Shell: AI uncovers critical RCE vulnerabilities affecting 70% of the Internet | Wiz BlogBoth involve Google Quantum AI, Gemini 4 Argon
- NewsAnthropic’s Transparency HubBoth involve Claude Sonnet 4.6
About
- Google Quantum AI
- Grok
- Palo Alto Networks Unit 42
- DeepSeek Flash
- Kimi K3
- Claude Sonnet 4.6
- Gemini 4 Argon
- Trajectory Labs