LLM Reliability

Real-Time Trust Scoring for LLM Data Extraction

A new arXiv paper explores how to score the trustworthiness of structured LLM outputs in real time, aiming to make data extraction systems more auditable, calibrated, and safer to deploy.

Deepfake Detection

Alethea Adds Reality Defender to Artemis Detection

Alethea has partnered with Reality Defender to bring deepfake detection into its Artemis platform, signaling stronger enterprise demand for integrated digital authenticity workflows.

AI Video

Script-to-Slide Grounding Advances AI Video Creation

A new paper introduces script-to-slide grounding for automatic instructional video generation, linking script sentences to slide objects so systems can produce more structured, context-aware educational videos.

Speech Deepfakes

Codec-Aware Model Targets Speech Deepfake Detection

A new paper proposes quantizer-aware hierarchical neural codec modeling for speech deepfake detection, targeting artifacts introduced by modern neural audio codecs used in synthetic speech pipelines.

AI detection

Benchmarking AI Text Detectors Under Attack

A new benchmark evaluates AI-generated text detectors across model families, domains, and adversarial rewrites, highlighting how fragile authenticity tools can be outside narrow test settings.

Synthetic Audio

Neuron-Level Emotion Control for Audio LLM Speech

A new arXiv paper explores neuron-level steering of emotional expression in speech-generative large audio-language models, pointing to finer control in synthetic voice systems and new questions for authenticity and misuse.

AI Training Data

Shutterstock Expands Licensed AI Training Content Library

Shutterstock broadens its licensed content offerings for AI model training, addressing the growing demand for legally cleared datasets in the synthetic media industry.

Deepfake Detection

Corporate Defense Strategies Against Deepfake Fraud

As generative AI fraud escalates, enterprises must deploy multi-layered defenses combining detection technology, employee training, and authentication protocols to protect corporate identity.

Samsung

Samsung Commits $73B for 2026 AI Chip Dominance Push

Samsung Electronics announces massive $73 billion investment for 2026, targeting leadership in AI semiconductors and high-bandwidth memory essential for generative AI workloads.

LLM evaluation

Noise-Response Calibration: New Protocol Fixes LLM Judge Bias

Researchers introduce a causal intervention protocol that calibrates LLM judges by measuring their response to noise perturbations, addressing systematic evaluation biases in AI assessment systems.

LLM evaluation

REAL: New RL Method Improves LLM Judge Accuracy for AI Evaluation

Researchers introduce REAL, a regression-aware reinforcement learning framework that trains LLM judges to produce more accurate evaluations by optimizing for numerical precision rather than classification.

Research

SCE-LITE-HQ: Smoother Visual Counterfactuals via Diffusion

New research combines StyleGAN and diffusion models to generate high-quality visual counterfactual explanations, advancing explainable AI while revealing techniques applicable to synthetic media generation.