Neura News

AI News

News reporting focused on AI and machine learning, covering the companies behind these technologies, their real-world applications, and the ethical concerns they raise. This includes areas like generative AI (large language models, text-to-image and video), speech tech, and predictive analytics.

Latest News

18 articles
Research

PlanFlip Attacks Target Multi-Agent LLM Planning Phase

New research from Yuhang Wang introduces PlanFlip, a framework of four prompt injection attacks targeting the planning phase of multi-agent LLM systems. The attacks exploit a single injection into the Planner agent's context to corrupt all downstream sub-tasks. Testing on nine frontier LLMs across 3,479 episodes revealed that stronger models like GPT-5 are more vulnerable, while reasoning-augmented models like DeepSeek-R1 show full resistance.

Jul 213 minNeura News
AI Tools

Disaggregated Prefill and Decode for LLM Inference on SageMaker HyperPod

AWS has introduced a method to separate the prefill and decode phases of large language model inference on SageMaker HyperPod. By running each phase on dedicated GPU pools connected via Elastic Fabric Adapter with RDMA, the approach eliminates interference that occurs when the phases share a single GPU. This results in more consistent per-token latency and independent scaling for long-context, high-concurrency workloads.

Jul 106 minNeura News
AI Tools

AWS details observability strategy for SageMaker AI LLM inference

AWS published a technical guide on comprehensive observability for large language models deployed on Amazon SageMaker AI. The approach separates monitoring into infrastructure quantity and LLM quality dimensions, using CloudWatch and Amazon Managed Grafana to visualize GPU utilization, cost, and response quality signals. The solution includes alert thresholds for metrics such as safety scores and composite quality scores, evaluated by an LLM-as-judge setup.

May 304 minNeura News
Developer

Critical Starlette bug imperils millions of AI agents

A vulnerability in the Starlette open source framework, which gets 325 million weekly downloads, puts millions of AI agents at risk. Tracked as CVE-2026-48710 and named BadHost, the flaw lets attackers bypass path-based authorization via a single character in the HTTP Host header. Affected packages include FastAPI, vLLM, and LiteLLM, and exploitation can expose sensitive data like clinical trial databases, email credentials, and cloud infrastructure details.

May 264 minNeura News
Research

New Study Reveals 'Constraint Decay' in LLM Backend Code Generation

Researchers at arXiv have identified a phenomenon they call 'constraint decay,' where large language model agents experience a sharp drop in performance as structural requirements accumulate in backend code generation tasks. The study tested agents across 80 greenfield tasks and 20 feature-implementation tasks spanning eight web frameworks, finding that agents succeed in minimal frameworks like Flask but struggle significantly in convention-heavy environments such as FastAPI and Django. Data-layer defects, including incorrect query composition and ORM runtime violations, were identified as the leading cause of errors.

May 243 minNeura News
AI Models

Hassabis sees AI singularity soon; LeCun says LLMs not intelligent

At Google I/O 2026, DeepMind co-founder Demis Hassabis declared humanity is entering the singularity, predicting AGI within five years. Yann LeCun of AMI Labs countered that current language models lack true intelligence, which he defines as solving new problems without prior training. Gemini co-lead Oriol Vinyals offered a middle ground, acknowledging models are strong in code and math but still missing experiential learning.

May 243 minNeura News