Events & Conferences
How we built Cedar with automated reasoning and differential testing
Cedar is a new authorization-policy language used by the Amazon Verified Permissions and AWS Verified Access managed services, and we recently released it publicly. Using Cedar, developers can write policies that specify fine-grained permissions for their applications. The applications then authorize access requests by calling Cedar’s authorization engine. Because Cedar policies are separate from application code, they can be independently authored, updated, analyzed, and audited.
We want to assure developers that Cedar’s authorization decisions will be correct. To provide that assurance, we follow a two-part process we call verification-guided development when we’re working on Cedar. First, we use automated reasoning to prove important correctness properties about formal models of Cedar’s components. Second, we use differential random testing to show that the models match the production code. In this blog post we present an overview of verification-guided development for Cedar.
A primer on Cedar
Cedar is a language for writing and enforcing authorization policies for custom applications. Cedar policies are expressed in syntax resembling natural language. They define who (the principal) can do what (the action) on what target (the resource) under which conditions (when)?
To see how Cedar works, consider a simple application, TinyTodo, designed for managing task lists. TinyTodo uses Cedar to control who can do what. Here is one of TinyTodo’s policies:
// policy 1
permit(principal, action, resource)
when {
resource has owner && resource.owner == principal
};
This policy states that any principal (a TinyTodo User) can perform any action on any resource (a TinyTodo List) as long as the resource’s creator, defined by its owner attribute, matches the requesting principal. Here’s another TinyTodo Cedar policy:
// policy 2
permit (
principal,
action == Action::"GetList",
resource
)
when {
principal in resource.editors || principal in resource.readers
};
This policy states that any principal can read the contents of a task list (Action::”GetList”) if that principal is in either the list’s readers group or its editors group. Here is a third policy:
// policy 3 forbid ( principal in Team::"interns", action == Action::"CreateList", resource == Application::"TinyTodo" );
This policy states that any principal who is an intern (in Team::”interns”) is forbidden from creating a new task list (Action::”CreateList”) using TinyTodo (Application::”TinyTodo”).
When the application needs to enforce access, as when a user of TinyTodo issues a command, it only needs to make a corresponding request to the Cedar authorization engine. The authorization engine evaluates the request in light of the Cedar policies and relevant application data. If it returns decision Allow, TinyTodo can proceed with the command. If it returns decision Deny, TinyTodo can report that the command is not permitted.
How do we build Cedar to be trustworthy?
Our work on Cedar uses a process we call verification-guided development to ensure that Cedar’s authorization engine makes the correct decisions. The process has two parts. First, we model Cedar’s authorization engine and validator in the Dafny verification-aware programming language. With Dafny, you can write code, and you can specify properties about what the code is meant to do under all circumstances. Using Dafny’s built-in automated-reasoning capabilities we have proved that the code satisfies a variety of safety and security properties.
Second, we use differential random testing (DRT) to confirm that Cedar’s production implementation, written in Rust, matches the Dafny model’s behavior. We generate millions of diverse inputs and feed them to both the Dafny model and the production code. If both versions always produce the same output, we have a high degree of confidence that the implementation matches the model.
Proving properties about Cedar authorization
Cedar’s authorization algorithm was designed to be secure by default, as exemplified by the following two properties:
- explicit permit — permission is granted only by individual permit policies and is not gained by error or default;
- forbid overrides permit — any applicable forbid policy always denies access, even if there is a permit policy that allows it.
With these properties, sets of policies are easier to understand. Policy authors know that permit policies are the only way access is granted, and forbid policies decline access regardless of whether it is explicitly permitted.
Given an authorization request, the Cedar authorization engine takes each Cedar policy and evaluates it after substituting the application request parameters into the principal, action and resource variables. For example, for the request principal= User::”Alice”, action=Action::”GetList”, and resource=List::”AliceList”, substituting for the variables in policy 1 would produce the expression List::”AliceList” has owner && List::”AliceList”.owner == User::”Alice”. If this expression evaluates to true, we say the request satisfies the policy. The authorization engine collects the satisfied forbid and permit policies into distinct sets and then makes its decision.
We model the authorization engine as a Dafny function and use Dafny’s automated-reasoning capabilities to state and prove the explicit-permit and forbid-overrides-permit properties. To see how this helps uncover mistakes, let’s consider a buggy version of the authorization engine:
function method isAuthorized(): Response { // BUGGY VERSION
var f := forbids();
var p := permits();
if f != {} then
Response(Deny, f)
else
Response(Allow, p)
}
The logic states that if any forbid policy is applicable (set f is not the empty set {}), the result should be Deny, thus overriding any applicable permit policies (in set p). Otherwise, the result is Allow. While this logic correctly reflects the desired forbid-overrides-permit property, it does not correctly capture explicit permit. Just because there are no applicable forbid policies doesn’t mean there are any applicable permit policies. We can see this by specifying and attempting to prove explicit permit in Dafny:
// A request is explicitly permitted when a permit policy is satisfied
predicate IsExplicitlyPermitted(request: Request, store: Store) {
exists p ::
p in store.policies.policies.Keys &&
store.policies.policies[p].effect == Permit &&
Authorizer(request, store).satisfied(p)
}
lemma AllowedIfExplicitlyPermitted(request: Request, store: Store)
ensures // A request is allowed if it is explicitly permitted
(Authorizer(request, store).isAuthorized().decision == Allow) ==>
IsExplicitlyPermitted(request, store)
{ ... }
A Dafny predicate is a function that takes arguments and returns a logical condition, and a Dafny lemma is a property to be proved. The IsExplicitlyPermitted predicate defines the condition that there is an applicable permit policy for the given request. The AllowedIfExplicitlyPermitted lemma states that a decision of Allow necessarily means the request was explicitly permitted. This lemma does not hold for the isAuthorized definition above; Dafny complains that A postcondition might not hold on this return path and points to the ensures clause.
Here is the corrected code:
function method isAuthorized(): Response {
var f := forbids();
var p := permits();
if f == {} && p != {} then
Response(Allow, p)
else
Response(Deny, f)
}
Now a response is Allow only if there are no applicable forbid policies, and there is at least one applicable permit policy. With this change, Dafny automatically proves AllowedIfExplicitlyPermitted. It also proves forbid overrides permit (not shown).
We have used the Cedar Dafny models to prove a variety of properties. Our most significant proof is that the Cedar validator, which confirms that Cedar policies are consistent with the application’s data model, is sound: if the validator accepts a policy, evaluating the policy should never result in certain classes of error. When carrying out this proof in Dafny, we found a number of subtle bugs in the validator’s design that we were able to correct.
We note that Dafny models are useful not just for automated reasoning but for manual reasoning, too. The Dafny code is much easier to read than the Rust implementation. As one measure of this, at the time of this writing the Dafny model for the authorizer has about one-sixth as many lines of code as the production code. Both Cedar users and tool implementers can refer to the Dafny models to quickly understand precise details about how Cedar works.
Differential random testing
Once we have proved properties about the Cedar Dafny model, we want to provide evidence that they hold for the production code, too, which we can do by using DRT to show that the model and the production code behave the same. Using the cargo fuzz random-testing framework, we generate millions of inputs — access requests, accompanying data, and policies — and send them to both the Dafny model engine and the Rust production engine. If the two versions agree on the decision, then all is well. If they disagree, then we have found a bug.
The main challenge with using DRT effectively is to ensure the necessary code coverage by generating useful and diverse inputs. Randomly generated policies are unlikely to mention the same groups and attributes chosen in randomly generated requests and data. As a result, pure random generation will miss a lot of core evaluation logic and overindex on error-handling code. To resolve this, we wrote several input generators, including ones that take care to generate policies, data, and requests that are consistent with one another, while also producing policies that use Cedar’s key language constructs. As of this writing, we run DRT for six hours nightly and execute on the order of 100 million total tests.
The use of DRT during Cedar’s development has discovered corner cases where there were discrepancies between the model and the production code, making it an important tool in our toolkit. For example, there was a bug in a Rust package we were using for IP address operations; the Dafny model exposed an issue in how the package was parsing IP addresses. Since the bug is in an external package, we fixed the problem within our code while we wait for the upstream fix. We also found subtle bugs in the Cedar policy parser, in how the authorizer handles missing application data, and how namespace prefixes on application data (e.g., TinyTodo::List::”AliceList”) are interpreted.
Learn more
In this post we have discussed the verification-guided development process we have followed for the Cedar authorization policy language. In this process, we model Cedar language components in the Dafny programming language and use Dafny’s automated-reasoning capabilities to prove properties about them. We check that the Cedar production code matches the Dafny model through differential random testing. This process has revealed several interesting bugs during development and has given us greater confidence that Cedar’s authorization engine makes correct decisions.
To learn more, you can check out the Cedar Dafny models and differential-testing code on GitHub. You can also learn more about Dafny on the Dafny website and the Cedar service on the Cedar website.
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
-
Funding & Business4 days ago
Dust hits $6M ARR helping enterprises build AI agents that actually do stuff instead of just talking
-
Funding & Business6 days ago
Europe’s Most Ambitious Startups Aren’t Becoming Global; They’re Starting That Way
