Events & Conferences
Long-form-video understanding and synthesis – Amazon Science
At this year’s Conference on Computer Vision and Pattern Recognition (CVPR), Prime Video presented four papers that indicate the broad range of cutting-edge problems we work on.
In one paper, “Movies2Scenes: Using movie metadata to learn scene representation“, we present a novel contrastive-learning approach that uses only commonly available movie metadata to learn a general-purpose scene representation. On a diverse set of tasks evaluated using multiple benchmark datasets, models that use our representations consistently outperform models using existing state-of-the-art representations.
Notably, our learned representation offers an average improvement of 7.9% on the seven classification tasks and 9.7% on the two regression tasks in the Long-Form Video Understanding (LVU) dataset. This effort is an important step toward the first foundation model for general-purpose movie understanding.
In another paper, “Selective structured state-spaces for long-form video understanding”, we expand on the recently proposed S4 model that employs a lightweight mask generator to adaptively select informative image tokens, resulting in more efficient and accurate modeling of long-term spatiotemporal dependencies in videos. Our approach is consistently more accurate than the previous state-of-the-art model, by as much as 9.6%, while reducing the memory footprint by 23%.
Similarly, our paper “Dynamic inference with grounding based vision and language models” explores the problem of computational redundancy in large vision-and-language models, addressing this challenge by dynamically skipping network layers, dropping input tokens, and fusing multimodal tokens, conditioned on the input image-text pair. Our results show that we can improve the run-time efficiency of the state-of-the-art models by up to 50% on multiple downstream tasks with an accuracy drop of only 0.3%.
Lastly, our paper “LEMaRT: Label-efficient masked region transform for image harmonization” addresses the problem of requiring large amounts of labeled data to train image harmonization models, which modify content from different source images so that they blend together better in composite images. To this end, our method automatically generates training data by simulating defects in appearance that image harmonization models are expected to remove. Our method outperforms previous state-of-the-art approaches by a margin of 0.4dB (mean square error improvement = ~9%) when it is fine-tuned on only 50% of the training data from one of the standard benchmarks (iHarmony4) and by 1.0 dB (MSE improvement = ~21%) when it is trained on the full training dataset.
Toward a foundation model for movie understanding
The term “foundation model” generally relates to (i) a single large model that is (ii) trained on large amounts of mostly unlabeled data and can (iii) drive a number of downstream tasks. While several general-purpose visual-and-textual foundation models exist (e.g., BERT, GPT-4, CLIP, DALL-E 2, etc.), no foundation model particularly geared for movie understanding has been proposed before our work.
This is partly because directly applying existing visual or textual foundation models for movie understanding has limited effectiveness, given the large domain gap between cinematic content and the web-crawled images and text used to train those models. Factors such as the inaccessibility of much large-scale cinematic content, the computational resources required to process it, and the lack of benchmark datasets for evaluation on downstream applications add to the challenge of building a foundation model for movie understanding.
To address these challenges, we proposed a novel model trained on over five million scenes automatically identified from thousands of movies and comprising more than 45 million frames. Our model does not require any manual annotations and relies only on commonly available movie-level information (genre, synopsis, etc.). The scene representations from our model can be applied to improve the performance of a diverse set of downstream tasks, which is a key step toward building a foundation model for movie understanding.
We use movie metadata to define a measure of movie similarity and use that similarity measure to identify data pairs for contrastive learning. In contrastive learning, a model is trained on both positive pairs — examples that are similar in the relevant way — and negative pairs. During training, the model learns to produce data representations that pull positive pairs together and push negative pairs apart.
Often, the positive pairs are created by augmenting existing examples — say, re-cropping them, reversing them, or re-coloring them. By instead using movies that are considered similar to each other (see below), we ensure that our positive scene-pairs are not only visually similar but also semantically coherent, providing us with a much richer set of geometric and thematic data augmentations that enhance the training objective beyond traditional augmentation approaches.
As can be seen in the video below, our learned scene representation is able to effectively put thematically similar scenes close to each other.
Qualitative examples of similar-scene pairs found using our approach.
In the examples below, we compare our representation with the commonly used CLIP visual representation for scene retrieval using place-labeled scenes in the Long-Form Video Understanding (LVU) dataset. Given a query scene, our representation can capture appearance as well as semantic concepts to retrieve similar scenes more effectively, while CLIP can capture only local appearance-based patterns. For overall retrieval precision on six categories of places, our representation offers a 22.7% improvement over CLIP.
Quantitatively, our learned representation exhibits an average improvement of 7.9% and 9.7% on the seven classification tasks and two regression tasks of the LVU dataset, respectively. Furthermore, using our newly collected MCD dataset in Prime Video, we compare our learned scene representation with state-of-the-art models pretrained on action recognition and image classification datasets. Our scene representation outperforms the alternatives by margins ranging from 3.8% to 50.9% across different models and tasks.
Reducing model complexity for long-form-video understanding
At Prime Video, we’re developing state-of-the-art AI models for cinematic-content understanding to facilitate a variety of downstream use cases. One of the key technical problems to this end is effective modeling of complex spatiotemporal dependencies, particularly in long-form videos such as movies and TV episodes.
Previously proposed convolutional and recurrent neural networks struggle to learn long-term dependencies. In part this is because of exploding or vanishing gradients — where cascading adjustments to model weights grow too small or too large — as information is incorporated over long durations. Vision transformers can use self-attention to address this challenge, attending to particular, prior frames of video when interpreting the current frame. But this is computationally expensive, as it requires pairwise computations between the current frame and its predecessors.
The recently proposed structured-state-space-sequence (S4) model, with its linear complexity, offers a promising direction in this space; however, we empirically demonstrate that treating all image tokens equally, as the S4 model does, can adversely affect a model’s efficiency and accuracy.
To address this challenge, we present a novel selective S4 (i.e., S5) model that employs a lightweight mask generator to adaptively select informative image tokens, resulting in more efficient and accurate modeling of long-term spatiotemporal dependencies in videos. Unlike previous methods, which used mask-based token reduction in transformers, our S5 model avoids the dense self-attention calculation by following the guidance of the momentum-updated S4 model. This enables our model to efficiently discard less informative tokens and adapt to various long-form-video-understanding tasks more effectively.
However, as is the case with most token reduction methods, the informative image tokens may be dropped incorrectly. To improve the robustness and the temporal horizon of our model, we propose a novel long-short masked contrastive-learning (LSMCL) approach that enables our model to predict longer temporal contexts using shorter input videos.
We present extensive comparative results using three challenging long-form video-understanding datasets (LVU, COIN, and Breakfast), demonstrating that our approach is consistently more accurate than the previous state-of-the-art S4 model, by as much as 9.6% on one dataset, with a memory footprint that’s 23% smaller.
Dynamic inference of multimodal models using reinforcement learning
The availability of transformer models operating over multiple data modalities as well as large-scale pretraining approaches has led to significant progress on joint image-and-language models. However, these models impose high computational costs and therefore offer low run-time efficiency, making them difficult to apply to Prime Video’s large catalogue.
Although approaches such as pruning, knowledge distillation, and quantization can help address this challenge, they can incur significant drops in accuracy (e.g., ≥ 1% at ≥ 50% model compression rates), as they are primarily designed for model-parameter reduction, not improving run-time efficiency.
To address this challenge, we propose a model that saves computation by dynamically skipping layers of a multimodal network; pruning input tokens from either the language backbone, the image backbone, or both; and fusing tokens from the separate backbones, conditioned on the input image-text pair.
Most multimodal transformer models include multihead self-attention and feed-forward network layers, which can be skipped for some inputs. Additionally, we remove redundant tokens at different levels of the backbones and fuse the image tokens with the language tokens in an adaptive manner. To learn policies for dynamic inference, we train agents using reinforcement learning.
Our results demonstrate that we can improve the run-time efficiency of the state-of-the-art models MDETR and GLIP by up to 50% on the tasks of referring-expression comprehension, segmentation, and visual question-answering, with a maximum accuracy drop of only 0.3%.
Improving label efficiency of image harmonization models
Image harmonization is an important component of the broader problem of image composition, where new images are created by extracting foreground regions from one image and transferring them to another image in a photorealistic manner.
The main technical challenge for image harmonization is the appearance mismatch between the foreground extracted from the source image and the background of the destination image. Image harmonization aims to adjust the appearance of the foreground to make it compatible with the background. However, training traditional models for image harmonization requires a large amount of labeled data, which is costly and time-consuming to obtain.
To address this challenge, we introduce a novel approach to pretraining image harmonization models, LEMaRT, which automatically generates training data by simulating the types of defects that image harmonization models are expected to remove. LEMaRT takes an image as input, selects a region in that image, and applies a set of appearance transformations to it. We use these modified images, along with the original images, to pretrain our image harmonization model. Furthermore, we introduce an image harmonization model, SwinIH, by retrofitting the previously proposed Swin Transformer with a combination of local and global self-attention mechanisms.
Pretraining our SwinIH model with our LEMaRT approach results in a new state of the art for image harmonization, while being label-efficient, i.e., consuming less annotated data for fine-tuning than existing methods. Notably, on the iHarmony4 dataset, SwinIH outperforms the state of the art, i.e., SCS-Co by a margin of 0.4 dB when it is fine-tuned on only 50% of the training data and by 1.0 dB when it is trained on the full training dataset.
Qualitative comparisons suggest that LEMaRT is better at color correction than prior methods, thanks to the pretraining process, during which LEMaRT learns the distribution of photorealistic images.
Have you ever worked is legacy code? Are you curious what it takes to modernize systems at a massive scale?
Pascal Hartig is joined on the latest Meta Tech Podcast by Elaine and Buping, two software engineers working on a bold project to rewrite the decades-old C code in one of Meta’s core messaging libraries in Rust. It’s an ambitious effort that will transform a central messaging library that is shared across Messenger, Facebook, Instagram, and Meta’s AR/VR platforms.
They discuss taking on a project of this scope – even without a background in Rust, how they’re approaching it, and what it means to optimize for ‘developer happiness.’
Download or listen to the episode below:
You can also find the episode wherever you get your podcasts, including:
The Meta Tech Podcast is a podcast, brought to you by Meta, where we highlight the work Meta’s engineers are doing at every level – from low-level frameworks to end-user features.
Send us feedback on Instagram, Threads, or X.
And if you’re interested in learning more about career opportunities at Meta visit the Meta Careers page.
Amazon Research Awards (ARA) provides unrestricted funds and AWS Promotional Credits to academic researchers investigating various research topics in multiple disciplines. This cycle, ARA received many excellent research proposals from across the world and today is publicly announcing 73 award recipients who represent 46 universities in 10 countries.
This announcement includes awards funded under five call for proposals during the fall 2024 cycle: AI for Information Security, Automated Reasoning, AWS AI, AWS Cryptography, and Sustainability. Proposals were reviewed for the quality of their scientific content and their potential to impact both the research community and society. Additionally, Amazon encourages the publication of research results, presentations of research at Amazon offices worldwide, and the release of related code under open-source licenses.
Recipients have access to more than 700 Amazon public datasets and can utilize AWS AI/ML services and tools through their AWS Promotional Credits. Recipients also are assigned an Amazon research contact who offers consultation and advice, along with opportunities to participate in Amazon events and training sessions.
“Automated Reasoning is an important area of research for Amazon, with potential applications across various features and applications to help improve security, reliability, and performance for our customers. Through the ARA program, we collaborate with leading academic researchers to explore challenges in this field,” said Robert Jones, senior principal scientist with the Cloud Automated Reasoning Group. “We were again impressed by the exceptional response to our Automated Reasoning call for proposals this year, receiving numerous high-quality submissions. Congratulations to the recipients! We’re excited to support their work and partner with them as they develop new science and technology in this important area.”
“At Amazon, we believe that solving the world’s toughest sustainability challenges benefits from both breakthrough scientific research and open and bold collaboration. Through programs like the Amazon Research Awards program, we aim to support academic research that could contribute to our understanding of these complex issues,” said Kommy Weldemariam, Director of Science and Innovation Sustainability. “The selected proposals represent innovative projects that we hope will help advance knowledge in this field, potentially benefiting customers, communities, and the environment.”
ARA funds proposals throughout the year in a variety of research areas. Applicants are encouraged to visit the ARA call for proposals page for more information or send an email to be notified of future open calls.
The tables below list, in alphabetical order by last name, fall 2024 cycle call-for-proposal recipients, sorted by research area.
AI for Information Security
| Recipient | University | Research title |
| Christopher Amato | Northeastern University | Multi-Agent Reinforcement Learning Cyber Defense for Securing Cloud Computing Platforms |
| Bernd Bischl | Ludwig Maximilian University of Munich | Improving Generative and Foundation Models Reliability via Uncertainty-awareness |
| Shiqing Ma | University Of Massachusetts Amherst | LLM and Domain Adaptation for Attack Detection |
| Alina Oprea | Northeastern University | Multi-Agent Reinforcement Learning Cyber Defense for Securing Cloud Computing Platforms |
| Roberto Perdisci | University of Georgia | ContextADBench: A Comprehensive Benchmark Suite for Contextual Anomaly Detection |
Automated Reasoning
| Recipient | University | Research title |
| Nada Amin | Harvard University | LLM-Augmented Semi-Automated Proofs for Interactive Verification |
| Suguman Bansal | Georgia Institute of Technology | Certified Inductive Generalization in Reinforcement Learning |
| Ioana Boureanu | University of Surrey | Phoebe+: An Automated-Reasoning Tool for Provable Privacy in Cryptographic Systems |
| Omar Haider Chowdhury | Stony Brook University | Restricter: An Automatic Tool for Authoring Amazon Cedar Access Control Policies with the Principle of Least Privilege |
| Stefan Ciobaca | Alexandru Ioan Cuza University | An Interactive Proof Mode for Dafny |
| João Ferreira | INESC-ID | Polyglot Automated Program Repair for Infrastructure as Code |
| Sicun Gao | University Of California, San Diego | Monte Carlo Trees with Conflict Models for Proof Search |
| Mirco Giacobbe | University of Birmingham | Neural Software Verification |
| Tobias Grosser | University of Cambridge | Synthesis-based Symbolic BitVector Simplification for Lean |
| Ronghui Gu | Columbia University | Scaling Formal Verification of Security Properties for Unmodified System Software |
| Alexey Ignatiev | Monash University | Huub: Next-Gen Lazy Clause Generation |
| Kenneth McMillan | University of Texas At Austin | Synthesis of Auxiliary Variables and Invariants for Distributed Protocol Verification |
| Alexandra Mendes | University of Porto | Overcoming Barriers to the Adoption of Verification-Aware Languages |
| Jason Nieh | Columbia University | Scaling Formal Verification of Security Properties for Unmodified System Software |
| Rohan Padhye | Carnegie Mellon University | Automated Synthesis and Evaluation of Property-Based Tests |
| Nadia Polikarpova | University Of California, San Diego | Discovering and Proving Critical System Properties with LLMs |
| Fortunat Rajaona | University of Surrey | Phoebe+: An Automated-Reasoning Tool for Provable Privacy in Cryptographic Systems |
| Subhajit Roy | Indian Institute of Technology Kanpur | Theorem Proving Modulo LLM |
| Gagandeep Singh | University of Illinois At Urbana–Champaign | Trustworthy LLM Systems using Formal Contracts |
| Scott Stoller | Stony Brook University | Restricter: An Automatic Tool for Authoring Amazon Cedar Access Control Policies with the Principle of Least Privilege |
| Peter Stuckey | Monash University | Huub: Next-Gen Lazy Clause Generation |
| Yulei Sui | University of New South Wales | Path-Sensitive Typestate Analysis through Sparse Abstract Execution |
| Nikos Vasilakis | Brown University | Semantics-Driven Static Analysis for the Unix/Linux Shell |
| Ping Wang | Stevens Institute of Technology | Leveraging Large Language Models for Reasoning Augmented Searching on Domain-specific NoSQL Database |
| John Wawrzynek | University of California, Berkeley | GPU-Accelerated High-Throughput SAT Sampling |
AWS AI
| Recipient | University | Research title |
| Panagiotis Adamopoulos | Emory University | Generative AI solutions for The Spillover Effect of Fraudulent Reviews on Product Recommendations |
| Vikram Adve | University of Illinois at Urbana–Champaign | Fellini: Differentiable ML Compiler for Full-Graph Optimization for LLM Models |
| Frances Arnold | California Institute of Technology | Closed-loop Generative Machine Learning for De Novo Enzyme Discovery and Optimization |
| Yonatan Bisk | Carnegie Mellon University | Useful, Safe, and Robust Multiturn Interactions with LLMs |
| Shiyu Chang | University of California, Santa Barbara | Cut the Crap: Advancing the Efficient Communication of Multi-Agent Systems via Spatial-Temporal Topology Design and KV Cache Sharing |
| Yuxin Chen | University of Pennsylvania | Provable Acceleration of Diffusion Models for Modern Generative AI |
| Tianlong Chen | University of North Carolina at Chapel Hill | Cut the Crap: Advancing the Efficient Communication of Multi-Agent Systems via Spatial-Temporal Topology Design and KV Cache Sharing |
| Mingyu Ding | University of North Carolina at Chapel Hill | Aligning Long Videos and Language as Long-Horizon World Models |
| Nikhil Garg | Cornell University | Market Design for Responsible Multi-agent LLMs |
| Jessica Hullman | Northwestern University | Human-Aligned Uncertainty Quantification in High Dimensions |
| Christopher Jermaine | Rice University | Fast, Trusted AI Using the EINSUMMABLE Compiler |
| Yunzhu Li | Columbia University | Physics-Informed Foundation Models Through Embodied Interactions |
| Pattie Maes | Massachusetts Institute of Technology | Understanding How LLM Agents Deviate from Human Choices |
| Sasa Misailovic | University of Illinois at Urbana–Champaign | Fellini: Differentiable ML Compiler for Full-Graph Optimization for LLM Models |
| Kristina Monakhova | Cornell University | Trustworthy extreme imaging for science using interpretable uncertainty quantification |
| Todd Mowry | Carnegie Mellon University | Efficient LLM Serving on Trainium via Kernel Generation |
| Min-hwan Oh | Seoul National University | Mutually Beneficial Interplay Between Selection Fairness and Context Diversity in Contextual Bandits |
| Patrick Rebeschini | University of Oxford | Optimal Regularization for LLM Alignment |
| Jose Renau | University of California, Santa Cruz | Verification Constrained Hardware Optimization using Intelligent Design Agentic Programming |
| Vilma Todri | Emory University | Generative AI solutions for The Spillover Effect of Fraudulent Reviews on Product Recommendations |
| Aravindan Vijayaraghavan | Northwestern University | Human-Aligned Uncertainty Quantification in High Dimensions |
| Wei Yang | University of Texas at Dallas | Optimizing RISC-V Compilers with RISC-LLM and Syntax Parsing |
| Huaxiu Yao | University of North Carolina at Chapel Hill | Aligning Long Videos and Language as Long-Horizon World Models |
| Amy Zhang | University of Washington | Tools for Governing AI Agent Autonomy |
| Ruqi Zhang | Purdue University | Efficient Test-time Alignment for Large Language Models and Large Multimodal Models |
| Zheng Zhang | Rutgers University-New Brunswick | AlphaQC: An AI-powered Quantum Circuit Optimizer and Denoiser |
AWS Cryptography
| Recipient | University | Research title |
| Alexandra Boldyreva | Georgia Institute of Technology | Quantifying Information Leakage in Searchable Encryption Protocols |
| Maria Eichlseder | Graz University of Technology, Austria | SALAD – Systematic Analysis of Lightweight Ascon-based Designs |
| Venkatesan Guruswami | University of California, Berkeley | Obfuscation, Proof Systems, and Secure Computation: A Research Program on Cryptography at the Simons Institute for the Theory of Computing |
| Joseph Jaeger | Georgia Institute of Technology | Analyzing Chat Encryption for Group Messaging |
| Aayush Jain | Carnegie Mellon | Large Scale Multiparty Silent Preprocessing for MPC from LPN |
| Huijia Lin | University of Washington | Large Scale Multiparty Silent Preprocessing for MPC from LPN |
| Hamed Nemati | KTH Royal Institute of Technology | Trustworthy Automatic Verification of Side-Channel Countermeasures for Binary Cryptographic Programs using the HoIBA libary |
| Karl Palmskog | KTH Royal Institute of Technology | Trustworthy Automatic Verification of Side-Channel Countermeasures for Binary Cryptographic Programs using the HoIBA libary |
| Chris Peikert | University of Michigan, Ann Arbor | Practical Third-Generation FHE and Bootstrapping |
| Dimitrios Skarlatos | Carnegie Mellon University | Scale-Out FHE LLMs on GPUs |
| Vinod Vaikuntanathan | Massachusetts Institute of Technology | Can Quantum Computers (Really) Factor? |
| Daniel Wichs | Northeastern University | Obfuscation, Proof Systems, and Secure Computation: A Research Program on Cryptography at the Simons Institute for the Theory of Computing |
| David Wu | University Of Texas At Austin | Fast Private Information Retrieval and More using Homomorphic Encryption |
Sustainability
| Recipient | University | Research title |
| Meeyoung Cha | Max Planck Institute | Forest-Blossom (Flossom): A New Framework for Sustaining Forest Biodiversity Through Outcome-Driven Remote Sensing Monitoring |
| Jingrui He | University of Illinois at Urbana–Champaign | Foundation Model Enabled Earth’s Ecosystem Monitoring |
| Pedro Lopes | University of Chicago | AI-powered Tools that Enable Engineers to Make & Re-make Sustainable Hardware |
| Cheng Yaw Low | Max Planck Institute | Forest-Blossom (Flossom): A New Framework for Sustaining Forest Biodiversity Through Outcome-Driven Remote Sensing Monitoring |
AI safety is a priority at Amazon. Our investment in safe, transparent, and responsible AI (RAI) includes collaboration with the global community and policymakers. We are members of and collaborate with organizations such as the Frontier Model Forum, the Partnership on AI, and other forums organized by government agencies such as the National Institute of Standards and Technology (NIST). Consistent with Amazon’s endorsement of the Korea Frontier AI Safety Commitments, we published our Frontier Model Safety Framework earlier this year.
During the development of the Nova Premier model, we conducted a comprehensive evaluation to assess its performance and safety. This included testing on both internal and public benchmarks and internal/automated and third-party red-teaming exercises. Once the final model was ready, we prioritized obtaining unbiased, third-party evaluations of the model’s robustness against RAI controls. In this post, we outline the key findings from these evaluations, demonstrating the strength of our testing approach and Amazon Premier’s standing as a safe model. Specifically, we cover our evaluations with two third-party evaluators: PRISM AI and ActiveFence.
Evaluation of Nova Premier against PRISM AI
PRISM Eval’s Behavior Elicitation Tool (BET) dynamically and systematically stress-tests AI models’ safety guardrails. The methodology focuses on measuring how many adversarial attempts (steps) it takes to get a model to generate harmful content across several key risk dimensions. The central metric is “steps to elicit” — the number of increasingly sophisticated prompting attempts required before a model generates an inappropriate response. A higher number of steps indicates stronger safety measures, as the model is more resistant to manipulation. The PRISM risk dimensions (inspired by the MLCommons AI Safety Benchmarks) include CBRNE weapons, violent crimes, non-violent crimes, defamation, and hate, amongst several others.
Using the BET Eval tool and its V1.0 metric, which is tailored toward non-reasoning models, we compared the recently released Nova models (Pro and Premier) to the latest models in the same class: Claude (3.5 v2 and 3.7 non-reasoning) and Llama4 Maverick, all available through Amazon Bedrock. PRISM BET conducts black-box evaluations (where model developers don’t have access to the test prompts) of models integrated with their API. The evaluation conducted with BET Eval MAX, PRISM’s most comprehensive/aggressive testing suite, revealed significant variations in safety against malicious instructions. Nova models demonstrated superior overall safety performance, with an average of 43 steps for Premier and 52 steps for Pro, compared to 37.7 for Claude 3.5 v2 and fewer than 12 steps for other models in the comparison set (namely, 9.9 for Claude3.7, 11.5 for Claude 3.7 thinking, and 6.5 for Maverick). This higher step count suggests that on average, Nova’s safety guardrails are more sophisticated and harder to circumvent through adversarial prompting. The figure below presents the number of steps per harm category evaluated through BET Eval MAX.
The PRISM evaluation provides valuable insights into the relative safety of different Amazon Bedrock models. Nova’s strong performance, particularly in hate speech and defamation resistance, represents meaningful progress in AI safety. However, the results also highlight the ongoing challenge of building truly robust safety measures into AI systems. As the field continues to evolve, frameworks like BET will play an increasingly important role in benchmarking and improving AI safety. As a part of this collaboration Nicolas Miailhe, CEO of PRISM Eval, said, “It’s incredibly rewarding for us to see Nova outperforming strong baselines using the BET Eval MAX; our aim is to build a long-term partnership toward safer-by-design models and to make BET available to various model providers.” Organizations deploying AI systems should carefully consider these safety metrics when selecting models for their applications.
Manual red teaming with ActiveFence
The AI safety & security company ActiveFence benchmarked Nova Premier on Bedrock on prompts distributed across Amazon’s eight core RAI categories. ActiveFence also evaluated Claude 3.7 (non-reasoning mode) and GPT 4.1 API on the same set. The flag rate on Nova Premier was lower than that on the other two models, indicating that Nova Premier is the safest of the three.
| Model | 3P Flag Rate [↓ is better] |
| Nova Premier | 12.0% |
| Sonnet 3.7 (non-reasoning) | 20.6% |
| GPT4.1 API | 22.4% |
“Our role is to think like an adversary but act in service of safety,” said Guy Paltieli from ActiveFence. “By conducting a blind stress test of Nova Premier under realistic threat scenarios, we helped evaluate its security posture in support of Amazon’s broader responsible-AI goals, ensuring the model could be deployed with greater confidence.”
These evaluations conducted with PRISM and ActiveFence give us confidence in the strength of our guardrails and our ability to protect our customers’ safety when they use our models. While these evaluations demonstrate strong safety performance, we recognize that AI safety is an ongoing challenge requiring continuous improvement. These assessments represent a point-in-time snapshot, and we remain committed to regular testing and enhancement of our safety measures. No AI system can guarantee perfect safety in all scenarios, which is why we maintain monitoring and response systems after deployment.
Acknowledgments: Vincent Ponzo, Elyssa Vincent
-
Funding & Business7 days agoKayak and Expedia race to build AI travel agents that turn social posts into itineraries
-
Jobs & Careers6 days ago
Mumbai-based Perplexity Alternative Has 60k+ Users Without Funding
-
Mergers & Acquisitions6 days ago
Donald Trump suggests US government review subsidies to Elon Musk’s companies
-
Funding & Business6 days ago
Rethinking Venture Capital’s Talent Pipeline
-
Jobs & Careers6 days ago
Why Agentic AI Isn’t Pure Hype (And What Skeptics Aren’t Seeing Yet)
-
Funding & Business4 days ago
Sakana AI’s TreeQuest: Deploy multi-model teams that outperform individual LLMs by 30%
-
Funding & Business7 days agoFrom chatbots to collaborators: How AI agents are reshaping enterprise work
-
Jobs & Careers6 days ago
Astrophel Aerospace Raises ₹6.84 Crore to Build Reusable Launch Vehicle
-
Jobs & Careers4 days ago
Ilya Sutskever Takes Over as CEO of Safe Superintelligence After Daniel Gross’s Exit
-
Funding & Business6 days ago
Europe’s Most Ambitious Startups Aren’t Becoming Global; They’re Starting That Way
