←
Neatsun Ziv186 posts
Neatsun Ziv avatar

Neatsun Ziv

@neatsun

Co-Founder & CEO of OX Security.

◷ Joined October 2009▧ 16 Photos and videos
Neatsun Ziv avatar
Neatsun Ziv@neatsun·Sep 16

The AI agents we're securing today aren't the ones we'll be securing tomorrow. We're watching agentic Darwinism play out in real time. Software that once followed predetermined instructions is evolving into agentic systems that can increasingly make decisions for themselves. Think of traditional software like a jellyfish: no brain, just predetermined paths and responses. Today's agents are more like frogs: we've given software a brain, but it's not fully in charge — yet. We don't know exactly what the next species will look like. But we can already see where the evolution is heading: toward systems where the brain plays a greater role in determining what happens, with more autonomy, more context, greater reach and more authority to act. That's what we need to build security for. Not the specific agents and architectures that exist today, but the capabilities that are likely to survive them. More on how to build security now for the agents that don't exist yet: ox.security/blog/how-do-you-… #AgenticAI #AISecurity #AgenticDarwinism

0:19
Neatsun Ziv avatar
Neatsun Ziv@neatsun·Aug 10

Great catching up with @ResilientCyber at Black Hat 🔥 My conversations this week keep coming back to the same question: What does security look like after the Mythos Age? Had a chance to show Chris what we're building at @OX__Security for the agentic age — the first AI-native application protection platform that governs the full agentic development lifecycle, from prompt to runtime. Always great to compare notes. Thanks for swinging by, Chris 💪

Post attachmentPost attachmentPost attachmentPost attachment
Neatsun Ziv avatar
Neatsun Ziv@neatsun·Aug 6

Fascinating conversation with @danielmiessler of Unsupervised Learning at Black Hat today. Daniel stopped by our booth to talk with me about security in the agentic age, and Field CTO Boaz Barzel walked him through @OX__Security’s solution: the first AI-native application protection platform that governs the full agentic development lifecycle, from prompt to runtime, in one system. Come say hi at Booth 5117 if you're around tomorrow 🤘 #AINAPP #BHUSA #CyberSecurity

Post attachmentPost attachmentPost attachmentPost attachment
Neatsun Ziv avatar
Neatsun Ziv@neatsun·Jul 28

The Human Age is over. Most people just haven't accepted it yet. We are now in the Mythos Age, where autonomous agents write the code, provision the infrastructure, and make production decisions in real time. There's no human in the loop anymore and there isn't time for one. That means no one owns code. You govern a derivative of it, if you govern anything at all. Problem is, the security industry is still built for a world where a person stood between the prompt and production — writing the code, approving the deploy, reviewing what shipped. CNAPP was built for that world. So was AppSec. So was vulnerability assessment. That person is gone, and so is that world. Your security team cannot see this layer. CNAPP protects containerized infrastructure. AppSec audits human-written code. Vulnerability assessment finds known flaws in static inventory. None of them address an agentic system making decisions before there's anything to audit at all. A new category of security is emerging as the answer to this: AINAPP — AI-native application protection, governing the full agentic lifecycle instead of one slice of it. We started building for this world before the category had a name, which is why @OX__Security is the first platform built to govern that entire cycle, prompt to runtime, end to end. Security used to have time to catch up. Now it doesn't. It has to move as fast as it is to prompt, or it isn't protecting anything at all.

Neatsun Ziv avatar
Neatsun Ziv@neatsun·Jun 12

5 days, 120 CISOs from around the globe, one Village. I was proud to take part in Team8 CISO Village Summit 2026, a gathering that reminded me why this community matters so much. We spent the week discussing AI, nation-state threats, insider risk, identity, exposure management, resilience, and the changing role of the CISO, but the real value was not only in the agenda. It was in the trust built among people facing the same hard questions from different angles. AI is reshaping cybersecurity, no doubt. It is changing how attackers move, how defenders respond, how software is built, and how risk is understood. But the strongest lesson from this week was that in the age of machines, the human network becomes even more important. The open conversations, the shared lessons, the uncomfortable questions, the willingness to say what is really happening inside our organizations, that is what turns a group of leaders into a Village. Thank you, Team8 CISO Village, for bringing together an extraordinary community of security leaders. In a world moving faster than any one organization can handle alone, the Village is not just a community. It is part of the defense. It takes a Village. #Team8CISOVillage @team8group

Post attachment
OX Security avatar
⟳ Neatsun Ziv retweeted
OX Security@OX__Security·Jun 10

🚨 JUST ANNOUNCED 🚨 VibeSecCon Returns on Tuesday - June 16th with: 🚀 @begimher | Sr. Security Engineer, @awscloud ⚡ @JamesBerthoty  | Founder & CEO, @latiotech 🚀 @pmungse | Head of Security, @Poshmarkapp ⚡ @chenxiwang | Managing General Partner, @rain_capital 🚀 @neatsun | CEO & Co-Founder, OX Security ⚡ @MosheTov | Security Research Team Leader 🚀 and more rockstars from @OX__Security, @mondaydotcom, @fastly, @AlphaSenseInc 🎟️ REGISTER TODAY 📍 ox.security/vibeseccon-retur…

0:29
Neatsun Ziv avatar
Neatsun Ziv@neatsun·Jun 1

did Mythos kill the CVE concept? AI didn't just speed up vulnerability discovery. It commoditized it. Give Mythos, GPT, or Llama a vulnerable function, and they'll find what's wrong every time. The hard part is no longer finding CVEs. It's knowing which ones matter. The CVE program wasn't built for this. MITRE's funding nearly lapsed earlier this year. Submissions are up 32%. The backlog keeps growing — and that's before AI floods the pipeline even further. Only 1.08% of findings correlate with actual risk, per OX's 2026 Application Security Benchmark. Security teams aren't drowning in vulnerabilities. They're drowning in noise. Here's what that misses: attackers don't exploit CVEs in isolation. They build chains. We found a clear example in DataEase — an open-source BI platform with 23,000+ GitHub stars. Four CVEs. Medium severity individually. Together: unauthenticated remote code execution. The chain only becomes visible when you understand the architecture. That's what the industry needs to shift toward — not individual findings, but Chains of Exploit. AI is already helping attackers build them. Security teams need the same lens. CVEs are dead. Long live the Mythos era. #CyberSecurity #AISecurity #AppSec #CVE #Anthropic @OX__Security

0:11
OX Security avatar
⟳ Neatsun Ziv retweeted
OX Security@OX__Security·May 19

🚨 300+ npm Packages Backdoored: 💥 Shai-Hulud Worm Hits 59M Downloads Preinstall runs on install → steals creds + spreads via GitHub. Impact: secrets exfiltrated at scale. Action: rotate keys, enable 2FA, assume compromise. FULL REPORT: ox.security/blog/the-antv-ec…

Moshe Siman Tov Bustan avatar
⟳ Neatsun Ziv retweeted
Moshe Siman Tov Bustan@MosheTov·May 17

TeamPCP: Attack of the Clones We detected the first ever out in the wild Shai-Hulud clone copying from the TeamPCP leaked source code last week. The Shai-Hulud variant is a part of a family of 4 different malwares by the same actor, including credential stealing logic, and also exfiltration of personal information (IP, Location, Hostname) AND a DDoS botnet server designed to attack servers directly from the victim's machine! The malwares are still live on npm, targeting developers via typo-squatting and Axios impersonation. For more details, and a full technical breakdown of all the four malicious packages: ox.security/blog/new-actors-…

Post attachment
OX Security avatar
⟳ Neatsun Ziv retweeted
OX Security@OX__Security·May 12

⚠️ BREAKING: #TeamPCP Source Code Hits GitHub Malware source code now in public repos Likely spread via compromised accounts/tokens Impact: fuels copycat attacks + rapid spread Action: audit repos, rotate tokens, enable 2FA ASAP FULL REPORT: ox.security/blog/shai-hulud-…

Post attachment
Neatsun Ziv avatar
Neatsun Ziv@neatsun·Apr 22

In the history of espionage, nothing ever stays contained. The moment Anthropic told the world that Mythos was the most powerful cyber model on earth, they turned themselves into the most desired target on earth. And things can get much worse than a simple "breach." Think about what Mythos is. It's a model designed to find vulnerabilities at a scale no human team can match. Now imagine an adversary didn't steal it -- they influenced it. Influenced it to miss a specific class of vulnerability. To treat specific malware as benign. To introduce a backdoor only inside federal systems. That model then gets deployed to defend enterprises, phones, critical infrastructure -- and per recent reporting, the NSA. Every one of those environments is now being secured through a lens the adversary shaped. That's not a breach. That's a generational compromise. A Discord group got in this time. A nation-state gets in next time. Mythos is a warning shot.

0:06
Neatsun Ziv avatar
Neatsun Ziv@neatsun·Apr 16

“By design” is not a defense against insecurity. Our research team at OX Security just uncovered a systemic vulnerability in Anthropic’s Model Context Protocol (MCP) impacting over 150+ million downloads. This isn’t a minor leak — it’s arbitrary command execution (RCE) across Python, TypeScript, Go, and more. When we disclosed it, the response was a familiar industry deflection: “It’s a feature, not a bug.” We’re seeing a dangerous trend where protocol architects shift the burden of security onto downstream developers — those with the least context and the most to lose. If a protocol sits at the most privileged junction of the AI stack, it must be secure by default. It’s a striking contradiction for Anthropic to launch Project Glasswing — a $100M initiative to find vulnerabilities in everyone else’s software — while declining to secure a flagrant RCE in its own foundation. Full findings here: ox.security/blog/the-mother-…

Kacper Szurek avatar
⟳ Neatsun Ziv retweeted
Kacper Szurek@KacperSzurek·Apr 3

Jak oszukano programistę projektu AXIOS: - Podszyto się pod firmę - Stworzono realistycznego slacka - Ofiarę zaproszono na spotkanie - Problem z rozmową - wymagana aktualizacja - Ofiara pod presją czasu instaluje malware - Konto NPM przejęte Widzieliście już coś takiego?

‼️ The axios lead maintainer has gone public on how he was socially engineered into installing the malware behind the npm supply chain attack. We have example images showing exactly how the attack was staged.

Post attachmentPost attachmentPost attachment
Neatsun Ziv avatar
Neatsun Ziv@neatsun·Apr 8

Anthropic just announced Project Glasswing and their most powerful model ever, Mythos. Everyone is celebrating the defensive angle. The coalition includes Amazon, Microsoft, Apple, Google, Cisco, CrowdStrike, Palo Alto Networks, NVIDIA, Broadcom, JPMorganChase, and the Linux Foundation — all signing up to use it to find vulnerabilities before attackers do. The 27-year-old OpenBSD bug found autonomously. The thousands of zero-days already patched. But nobody is asking the harder question. Anthropic themselves — and multiple outlets covering the launch — are describing Mythos as a general purpose reasoning and coding model. Not a dedicated cybersecurity tool. Its cyber capabilities are described as a byproduct of its raw coding and reasoning power, not something they specifically trained for. If that's true, it means something uncomfortable: Every frontier model that follows — from every lab, in every country, open source or closed — will eventually reach the same capability. You don't need Mythos. You need a model that's good enough at reasoning and code. And that bar gets crossed by more players every quarter. Which means the exploit code that gets generated won't all flow into public databases. Some will be government held. Some will be in the hands of threat actors who got there before the defenders did. Some will sit inside private bug bounty programs — programs that are already signaling they cannot handle the volume that agentic pentesting tools are generating today. And the two programs the industry relies on to understand what's being exploited in the wild — EPSS and KEV — are running on reduced DHS budgets. They were never designed to track what isn't publicly disclosed. They track what surfaces. What surfaces is becoming a smaller and smaller fraction of what exists. The gap between "this exploit exists" and "this exploit is public" is about to get much wider. Not smaller. So here's the question every security team needs to sit with: Your vulnerability prioritization logic — the model that tells you what to fix first — is it built for the world where exploit code is scarce and public disclosure is the signal? Or is it built for the world where exploit code is abundant, privately held, and the absence of a public PoC tells you almost nothing? If you have an RCE in your backlog with no known exploit and you're using that as justification to deprioritize it — don't. That exploit may already exist. You just don't have visibility into it.