Capital One Tech @CapitalOneTech

We strongly support the call for collective action on cyber defense. Together, we can make the digital infrastructure we all depend on more secure.

Greg Brockman @gdb

4 weeks ago

An open letter for a global surge in cyber defense, signed by over 100 organizations including Anthropic, AWS, Google, Microsoft, OpenAI, and Oracle. x.com/i/article/2093…

414 691 4K 1.4M 2K

@IEEESpectrum Read the full article on IEEE Spectrum: spectrum.ieee.org/capital-one-sc…

Instead of relying on third-party API wrappers, we build custom, proprietary AI models trained on our modern data ecosystem. Our Chief Scientist Prem Natarajan recently joined @IEEESpectrum to discuss how our research team advances model customization for complex financial environments.

Read the full survey paper to explore the red teaming frameworks and methodologies: capitalone.science/publications/r…

Defending LLMs requires a proactive offensive strategy. That’s why Capital One researchers published an end-to-end overview of LLM red teaming, which involves proactively attacking models to identify vulnerabilities. Their work maps out attack methods, software packages, and evaluation metrics for practical applications.

@Google @AnthropicAI Read our full blog on building an AI-ready workforce: capitalone.com/tech/ai/advanc…

As part of our commitment to AI building AI fluency at scale, we recently brought together over 10,000 of our engineers for a month of hands-on training to master agentic coding workflows with partners like @Google and @AnthropicAI.

Check out the full paper to see the architecture and benchmark results on LongBench and ∞Bench: capitalone.science/publications/f…

Complex RAG queries often force a choice between missing context and bloated prompts. Capital One researchers introduced FB-RAG, a training-free framework using lightweight LLMs to sample future outputs and guide the final generator—cutting latency up to 48%.

Read more from Milind Naphade, SVP, AI Foundations, on our science-first approach to Enterprise AI via Forbes: forbes.com/sites/capitalo…

At Capital One, we’re building custom AI stack layers and task-specialized models to run AI reliably at scale. By pairing open-source models with proprietary data and breaking down workflows, we get better latency, compute efficiency, and deterministic logic.

Read the full paper to explore our benchmark results and architecture on our science site: capitalone.science/publications/z…

Capital One researchers developed APT (Adversarially Pre-trained Transformer) for zero-shot tabular prediction. APT uses adversarial synthetic data agents and a mixture block architecture to handle arbitrary class counts with sub-second runtimes.

AI security improves when organizations share research, tools, and real-world experience. We’re joining other industry leaders, including @NVIDIA, in the Open Secure AI Alliance to help organizations identify and address software vulnerabilities and strengthen critical systems. Learn more: blogs.nvidia.com/blog/open-secu…

NVIDIA @nvidia

2 months ago

AI security advances when the industry builds in the open, together. We're introducing the Open Secure AI Alliance with industry leaders to develop new techniques and tools to safeguard software and agents. By sharing models, tooling and research in the open, we can broaden the

603 1K 9K 5.9M 1K

Explore our research and learn more about upcoming AI World Tour stops capitalone.com/tech/ai-world-…

Earlier this month, we wrapped up our July tour stops at #ICML2026 and #ACL2026NLP! Our teams shared peer-reviewed research on LLM reasoning and agentic safety with a focus on how we build responsible and scalable foundation model architectures.

I’m excited to announce that @CapitalOne is open-sourcing VulnHunter, an advanced agentic AI security tool we built to help the entire developer ecosystem find and fix software vulnerabilities. The digital threat landscape is moving faster than human defenders can keep pace. Because modern supply chains are deeply interconnected, a single vulnerability can ripple across thousands of enterprises simultaneously. No single organization can solve this challenge alone, and stakes are only rising for security teams to fight AI-enabled threats with equally capable AI-driven defenses to protect our digital environments. We designed VulnHunter with a developer-first mindset to solve a massive industry pain point: overwhelming false positives that create friction and slow down daily workflows. VulnHunter represents a new approach through three primary technical innovations:  -Falsification Engine: It actively tries to disprove its own findings to virtually eliminate false alarms. -Attacker-First Analysis: It simulates a hacker's journey, tracing entry points to map real attack paths. -Evidence-Backed Repair: It generates the exact code changes needed to fix verified defects. I’m incredibly proud of our internal team for exemplifying Capital One’s culture of innovation and invention. I’m also appreciative of our partnership with @AnthropicAI. VulnHunter’s advanced reasoning workflow is optimized to run using the exceptional capabilities of Claude Opus 4.8 and the Claude Code environment. It’s a noteworthy milestone in our collaboration, and I can’t wait to see how we’ll continue to create state-of-the-art capabilities together. At Capital One, we hope the release of VulnHunter enables the broader tech and security community to inspect the workflow, challenge its assumptions, and contribute improvements. Collectively using and contributing to these solutions can help advance the betterment of the systems we all build and the customers who depend on us to keep their data safe. Get the details here: i.capitalone.com/JyA3kUkDu

@AnthropicAI Dive into the full project and code on our GitHub repository: github.com/capitalone/vul…

Read more about the project on our tech blog. It is optimized for @AnthropicAI’s Claude Opus 4.8 and runs directly within Claude Code: i.capitalone.com/GWUe2ebrE

Traditional code scanners can flood developers with false positives. That’s why we built VulnHunter, an open-source code security tool with an agentic reasoning workflow. It leverages a unique falsification engine to challenge its own findings and simulate real attacker paths.