Ayal Yogev

We're excited to introduce our latest investment, Oryo.io - the security layer for agentic AI. Prashanth R. and Swaroop S., veterans of Okta and Wiz, are building agentic visibility and a trust layer on the MCP protocol for enterprise autonomous agents. Like Waymo’s cars today, agentic AI across enterprises will soon ingest data, make decisions, and take action. As adoption rises, the problem shifts from capability to control. CISOs manage identity and access for humans today. Okta anchors that trust layer for non-humans tomorrow. After our Agentic AI conference in Feb 2025 led by Anand Rao, we came with prepared minds to our first call with Oryo. Michael Marmor led diligence and is working closely with the founders as they build their MVP. We wrote the first check into Oryo. Village Global (Reid Hoffman) and Seven Hill Ventures (Michael Coates) followed soon after. Perhaps Nikesh Arora should take notice :) Venkat Raju, JB Michel, Rahul Harikumar, Quentin Staes-Polet Josh Cunningham

117

9 Comments

2 lines of code to detect audio, video, and still image deep fakes, via a seamless, high-scale API 2 lines of code to save your business or government agency from disaster 👇

9

Real time detection of cyber attacks on the application layer has historically been a major blind spot. Using runtime telemetry, the Contrast Security Graph enables SecOps to detect and stop breaches as they happen. From detection to deployable fix in mere minutes via Contrast's context-aware AI. #ADR #AICodeSecurity #RuntimeSecurity #AppSec

16

2 Comments

$50M in 10 months - here’s how I think Echo got here: #1 We stayed in our lane. After Argon, we didn’t chase a new market for our second start up. We targeted a problem in the world of container vulnerabilities that we’d seen firsthand with every Fortune 500 company we worked with. #2 We prioritized partners from day one. We understood that ease of use is essential for cloud security solutions, so we started with our partners. We made sure that echo integrates with all of the major container scanners and registries so our customers wouldn’t have to change the way they work today. #3 We hired experts in our space. We brought in engineering, product, and sales leaders who didn’t require training wheels in container security. They built the scanners and tools we now integrate with, bringing tons of expertise with them. #4 We made a deliberate choice to stand out. While deep cloud security experience mattered for building the product, we intentionally brought in marketing and design talent from outside the industry so we wouldn’t sound like everyone else. This Series A round is an exciting milestone, but it’s just the beginning. Big big things are coming, so stay tuned. N47 Notable Capital Hyperwise Ventures SentinelOne SVCI - Silicon Valley CISO Investments Eylam Milner

565

173 Comments

OpenAI & Paradigm just dropped EVMBench — and every security professional needs to pay attention. EVMBench is a new open-source benchmark evaluating AI agents' ability to detect, patch, and exploit smart contract vulnerabilities — built from 120 curated vulns across 40 real-world audits, including scenarios from Stripe's Tempo L1. This is going to accelerate blockchain security in a meaningful way. Having a standardized, reproducible framework for measuring AI capability against on-chain code is something the ecosystem has desperately needed. The data alone is already revealing: GPT-5.3-Codex scores 72.2% in exploit mode — up from 31.9% just six months ago with GPT-5. That rate of improvement is staggering. But here's what my traditional cybersecurity colleagues need to understand — because your mental model is wrong for this domain: Blockchain is not your patch-Tuesday world. In enterprise security, the lifecycle is detect → patch → deploy. You find a vuln, you push a fix, you move on. That model breaks completely with smart contracts. Most deployed contracts are immutable. Once on-chain, the code cannot be changed. No patch. No hotfix. No emergency update. The real-world attack sequence for blockchain will be: Detect → Exploit → Profit — long before any remediation is possible, if remediation is even an option. EVMBench's own results confirm the asymmetry: AI is already significantly better at exploiting contracts than detecting or patching the vulnerabilities that enable it. For legacy DeFi infrastructure securing $100B+ in assets, this has massive implications: → Contracts deployed years ago with no upgrade path now face AI-powered adversaries that didn't exist when that code was written → The cost of exploit discovery just dropped by orders of magnitude → Pre-deployment auditing alone is no longer sufficient — it needs to be continuous, adversarial, and AI-augmented → Protocols without proxy patterns, timelocks, or circuit breakers need to reassess their risk posture immediately The value EVMBench brings is giving the ecosystem a common language and measurement framework to quantify this threat and build defenses against it. This is exactly the kind of tooling that bridges the gap between AI capability research and real-world blockchain security operations. https://lnkd.in/eh29DRHy

15

Many companies cut the first security check long before they build the first security strategy, and the gap isn’t cheap. Our new Holtium insight shows how to spot that lag early and quantifies what it costs. Worth a 3-minute read if you’re on a growth path. 👇

17

Here’s a pattern I’m seeing more often: More sites. More assessments. More reporting expectations. Same headcount. Security teams are being asked to scale output without scaling structure. So what happens? Assessments become episodic. Reporting takes too long. Prioritization becomes subjective. And leaders spend more time translating risk than reducing it. This isn’t a capability issue. It’s an architecture issue. At some point, physical security has to operate with the same operational discipline as finance and IT. Otherwise it stays in permanent catch-up mode. For security people overseeing medium to large portfolios (20+ sites): What’s currently your biggest bottleneck volume, visibility, or validation? And why do you think this is?

5

1 Comment

📖 https://lnkd.in/g5YDnuCw Is ASLR giving us a false sense of security? Side channel attacks — like timing analysis, cache probing, and page fault observations — are quietly undermining memory randomization. Once ASLR is bypassed, even basic memory exploits become deadly again. The takeaway? Side channels aren’t just theoretical—they’re subtle, powerful, and growing. With each hardware or OS innovation, new side channel paths emerge. 💡 We need layered defenses: dynamic ASLR, low-resolution timers, stricter sandboxing, and better detection of recon patterns. Are side channels the next major frontier in memory exploitation? What should modern threat models include that they often don’t? #Cybersecurity #InfoSec #ASLR #SideChannel #MemorySecurity #ThreatModeling

3

Cyberattacks are getting smarter. Are your defenses keeping up? AI can help you: * Spot threats in real time * Predict risks before they spread * Cut through alert noise * Respond faster * Meet compliance needs But tools alone aren’t enough. You need the right strategy. See how AI + expert guidance can improve your cybersecurity. Read the full post: https://hubs.la/Q03szv3J0 #Cybersecurity #AI #ArtificialIntelligence #Compliance

20