Security Researchers: Digital Fighters Series
Upwind: “By thinking like an adversary the research team stays ahead of the curve”
Uriel Yochpaz, Vulnerability Research Lead at Upwind, discusses why defense requires an attacker's mindset and how healthy rivalry drives the industry forward, as part of CTech’s Security Researchers series.
“In many ways, a great deal of what the company builds starts with the research team,” says Uriel Yochpaz, Vulnerability Research Lead at cloud and AI security company Upwind. “By continuously thinking like an adversary and uncovering new attack techniques, the research team ensures the company stays ahead of the curve and always has a finger on the pulse of the evolving threat landscape.”
Within Israel’s cyber companies are small, highly specialized teams trained to think like attackers, find vulnerabilities and stay ahead of a threat landscape increasingly accelerated by AI. In this series, we meet the individuals and teams who make up this frontline of cyber: the digital fighters.
Yochpaz’s career in security research began as a Red Teamer in the IDF's J6 & Cyber Defense Directorate before moving to Unit 8200 to lead complex network operations as an Offensive Security Researcher. Following a stint at Upstream Security, Yochpaz now helps lead a department of approximately 30 researchers distributed across specialized groups at Upwind.
You can read the entire interview below.
ID Card
Company name: Upwind Security
Founders: Amiram Shachar, Tal Zur, Liran Polak, and Lavi Ferdman
Year of founding: 2022
Current number of employees: ~550
Company Description:
Upwind is a next-generation cloud and AI security company that has pioneered a runtime-first approach to protecting cloud & AI infrastructure and applications. Its unified Cloud-Native Application Protection Platform (CNAPP) uses runtime context from eBPF sensors to prioritize real risks and drive AI-powered response, consolidating capabilities that traditionally required multiple separate tools into a single platform.
Upwind connects build-time and runtime intelligence, allowing security, engineering, and DevOps teams to cut through the noise and focus on the threats that are actually exploitable in production. Upwind's vision is to deliver real-time, signal-over-noise cloud security built for the demands of the modern real-time era.
About Upwind’s Security Research Team:
The Research Department consists of approximately 30 researchers distributed across several specialized groups. The Platform Research team focuses on securing cloud and AI infrastructure, with the objective of building the world's leading Cloud Security Posture Management (CSPM) solution. The Threat and Runtime Research department specializes in detecting and tracking adversaries, as well as advancing knowledge in threat intelligence.
Additionally, the AI Department develops innovations to revolutionize cybersecurity utilizing advanced AI capabilities and Upwind AI models. Finally, the Offensive Department designs offensive security tools to validate security issues identified by other departments and conducts proactive vulnerability research to discover flaws ahead of potential threat actors.
What is your background in cyber, and what led you to specialize in security research?
My career in offensive security began in the IDF, where I served in the J6 & Cyber Defense Directorate's Unit as a Red Teamer before being promoted to Team Lead, overseeing a team of security researchers. I then joined Unit 8200 as an Offensive Security Researcher, leading complex network operations and developing custom offensive tooling.
Following my military service, I joined Upstream Security's CTO office, where I uncovered vulnerabilities across client-facing platforms and led the design and integration of GenAI capabilities into the company's API security product.
I now lead the vulnerability research team at Upwind Security. Throughout my career, I have discovered multiple CVEs and zero-day vulnerabilities in widely-used products and technologies, hold the OSCP certification, won a competitive cyber CTF, and hold a B.Sc. in Computer Science from the Open University.
What does your security research team look like in action?
Once we've selected a target for research (a specific technology or platform) our first step is to understand its legitimate, intended use and the full range of functionality it offers. From there, we map the attack surface and analyze the existing threat model. We then adopt an attacker's mindset for each attack surface, looking for ways to abuse existing capabilities in order to achieve outcomes that shouldn't be possible through legitimate use.
We prioritize research topics based on several factors: relevance to our customers, so our findings deliver direct value both to them and to the company's product; broad real-world adoption, so that our research has the widest possible impact; the potential severity of a vulnerability, focusing on products where a security flaw could lead to significant or dangerous consequences; and alignment with the AI and cloud domains.
We always begin by setting up a dedicated lab environment, which lets us work efficiently and securely in isolation. Our toolkit includes proxy tools for intercepting and manipulating network traffic, AI-powered tooling and MCP servers, and disassemblers for lower-level analysis.
Everyone works across similar topics, but each team member has their own niche which allows them to bring specialized depth to the rest of the team. We're especially interested in AI, including the various libraries and components used within AI systems, and of course cloud, both of which are the company's core domains.
How does the research team influence your company at large?
In many ways, a great deal of what the company builds starts with the research team, which is precisely what makes it so significant. High-quality, comprehensive research gives the product team the insights and conclusions they need to develop new capabilities and push the product forward. This is where the attacker mindset becomes especially valuable: by continuously thinking like an adversary and uncovering new attack techniques, the research team ensures the company stays ahead of the curve and always has a finger on the pulse of the evolving threat landscape.
Beyond product impact, rigorous research and the discovery of complex, sophisticated vulnerabilities position the company as a deeply technically credible organization — one that doesn't just sell a product, but genuinely understands the threat environment at an expert level. That reputation translates directly into trust, both with customers and within the broader security community.
What has been your team’s most significant security discovery to date?
One of the team's most significant discoveries was a novel attack technique that bypasses the AWS Data Perimeter — widely considered one of the strongest security boundaries available in cloud environments — using Amazon Bedrock AgentCore, AWS's managed service for deploying AI agents.
The AWS Data Perimeter is a layered defense model that simultaneously enforces trusted identity, trusted resources, and expected network origin for every API call. The research demonstrated that an attacker with a single code execution foothold inside a protected environment could establish a full bidirectional Command-and-Control (C2) channel by abusing two legitimate behaviors of the AgentCore service.
The first channel enabled data exfiltration: when an AgentCore Runtime is configured with a JWT discovery URL, the service issues a server-side HTTP request to validate it, originating from AWS infrastructure itself, not from the customer's VPC. By encoding sensitive data into the URL path, an attacker could exfiltrate information with each runtime update.
The second channel enabled command infiltration: an unauthenticated API endpoint — GetRuntimeProtectedResourceMetadata — was accessible through VPC endpoints even under strict deny-all policies, and produced no CloudTrail logs.
Together, the two channels formed a complete bidirectional C2 using only legitimate AWS service functionality. The research was presented at fwd:cloudsec North America 2026, and following responsible disclosure, AWS deployed a fix for the infiltration channel across all production regions. The exfiltration channel, which AWS considers standard OIDC behavior, remains open as AWS evaluates further defense-in-depth improvements.
Who or what is your 'Moby Dick'?
My "Moby Dick" is a cross-tenant vulnerability in a cloud environment. The impact of such a flaw is enormous. It could compromise every customer of the affected cloud provider, breaking the fundamental isolation that the entire model relies on.
A close second is unauthenticated RCE: the ability to achieve code execution with no prior authentication and no pre-existing foothold. For obvious reasons, that ranks among the most critical findings imaginable.
How would you characterize the competition between research teams today?
At the team level, I think there's a strong sense of camaraderie and a genuine willingness to help one another, because one researcher's success is the whole company's success. Looking outward at other companies, there's also a great deal of knowledge sharing and humility.
Cybersecurity research is a field people love to discuss openly and with an open mind, in order to enrich everyone's collective knowledge. And, of course, there's the credit that rightly goes to whoever did the work.
I think this subtle competitiveness is a healthy thing: seeing which company uncovers the most impactful vulnerabilities pushes researchers to raise their game.
There's an old Talmudic saying that captures this well: "The envy of scholars increases wisdom" (kinat sofrim tarbeh chochma) — the idea that rivalry among the learned drives everyone to greater heights.
What is your take on the future of the human security researcher?
I'd answer this from two angles. The first is the power AI gives the researcher, which is exponential. Problems that used to take days of effort to solve, or even just to attempt, now take seconds or minutes. A researcher can discover vulnerabilities far faster and weaponize them in seconds. That said, I still believe an experienced researcher is essential to guide and refine the AI to avoid drowning in mountains of irrelevant information and instead pinpoint exactly what's relevant to the research.
The second angle is that AI opens up an entirely new domain that demands dedicated security attention in its own right: from prompt injection to the security challenges of MCP.
