Check Point introduced PromptGuardX, a file-layer defense that scans PDFs for hidden or obfuscated prompt-injection instructions before extracted content reaches an agent, complementing runtime monitoring and action controls. The product is integrated into its Threat Emulation pipeline for supported gateways; efficacy claims are vendor-described, with independent evaluation not found. Source: https://blog.checkpoint.com/ai-security/check-point-promptguardx-securing-the-files-ai-agents-trust/
A Carnegie Mellon-led paper surfaced on Hugging Face proposing Selection-based Structured Reasoning (SSR): multimodal agents select from reusable reasoning candidates instead of generating free-form reasoning before every action. On seven search benchmarks using 2B/4B models, the authors report over 90% lower per-turn reasoning latency and 28–54% lower total inference latency without losing comparable task success; results are preliminary paper claims. Sources: https://arxiv.org/abs/2610.01892 ; https://huggingface.co/papers/2610.01892 ; https://zfy0314.github.io/ssr-webpage/