AI Briefing
Amazon Science news
The latest AI and developer news from Amazon Science, with short summaries.
Visit website →Customized Amazon Nova Models Improve Molecular Property Prediction for Drug Discovery
Customized Amazon Nova models have improved the performance of molecular property prediction for drug discovery.
AWS and Johns Hopkins Engineering Release Large-Scale Dataset for AI/ML Antibody Design
AWS and JHU have released a large-scale benchmark for evaluating AI antibody design.
How Amazon Achieves Global-Scale Vulnerability Detection with Agentic AI
RuleForge automated security detection rule generation and validation 336% faster.
How Amazon Verifies and Optimizes Post-Quantum Cryptography
Amazon achieved a high-performance, high-assurance implementation of ML-KEM through formal verification and optimization.
Improving Quality and Robustness of LLM-based Text-to-Speech Systems
LoRA, CFG, and chain-of-thought reduced accent leakage, hallucination, and truncation in LLM-based TTS.
Formally Verified AES-XTS: The First AES Algorithm to Join s2n-bignum
Arm64 AES-256-XTS was verified with HOL Light and added to s2n-bignum.
Optimizing LoRA Target Module Selection for Efficient Fine-Tuning
Based on Nova 2.0 Lite, attaching LoRA to only **o_proj** gave the best balance of efficiency and accuracy.
Agentic AI that heals systems that can't be replaced
Agentic AI makes it possible to handle old legacy systems stably without replacing them.
Designing When AI Agents Should Step Back
The core of agentic UX is not autonomy itself, but coordinating when to step in and when to step back.
How AI Is Changing the Nature of Mathematical Research
AI speeds up mathematical proofs and paper writing, but it is shaking up education and peer review.
Intelligence Is Time, Not Parameter Count
An LLM's intelligence depends less on scale than on inference time and the algorithmic structure it has learned.
Why a 12-Year-Old Forecasting Paper Still Holds Up
**EMBERS** is an old model, but its way of evaluating future forecasts and fusing warnings is still valid today.
How Academic Collaboration Turns Into Real Security for Amazon Customers
Joint research with Stanford grew into cvc5, which now strengthens AWS security.
Amazon Nova AI Challenge returns for a second season with Nova Forge access
The Amazon Nova AI Challenge is reopening its AI coding agent competition with 10 university teams.
Amazon Connects with the AI Community Through Building, Research, and Shared Learning
Amazon has launched a hackathon and a university research challenge together, centered on Nova.
A Decade of Innovation in NFL Next Gen Stats
Starting with RFID and ML, Next Gen Stats has become core infrastructure for the NFL's gameplay and safety.
Customizing multi-turn AI agents with reinforcement learning
AWS significantly boosted multi-turn agent performance using online RL with a small amount of data.
Fine-Tuning Vision-Language Models on Memory-Constrained Devices
SharpZO fine-tunes VLMs using only forward passes, improving both accuracy and speed.
The Invisible Work of Building Trustworthy AI Agents
The core of AI agents lies not in flashy features but in the reliability of scrolling and clicking.
Top 10 Most-Viewed Publications of 2025
Amazon Science's most-viewed publications of 2025 were led by Nova, safety, and robotics.