Frontier AI can now probe an internet-facing app, try dozens of exploit paths, and chain together flaws that look minor on their own, all in minutes. Scanning and patching can’t keep up with that pace, and neither can a pentest that runs once or twice a year.
Today, Zafran is announcing Offensive Validation, the newest part of Zafran’s Autonomous Exposure Prevention platform. Zafran already shows what is exploitable by layering environmental context with the security controls you have in place. Offensive Validation adds proof. AI agents use that context to test your environment the way a real attacker would, confirm which known findings an attacker can actually use, and uncover unknown exposures that scanners never flag. With it, Zafran completes the full cycle of exposure management: assess, validate, fix, and verify. Each proven exposure is routed agentically to the relevant owner with a detailed fix that can be tested and verified in one click. Zafran relies on an enterprise-ready Exposure Graph, enriched with your environment’s context, to reduce the number of fixes and keep production safe and running.
Agentic attacks are outpacing the fix cycle
On November 13, the FBI issued a FLASH alert about ARTEX, an open source penetration testing framework run by AI agents. It was built to help security teams test their own defenses. Attackers have turned it into a low-cost way to attack at scale. Since at least August, they have used it against finance, government, technology, healthcare, and other sectors, in some cases going from picking a target to privileged access in under three hours. ARTEX goes after familiar weak spots, such as exposed admin panels, weak credentials, unpatched apps, and leaked secrets, at machine speed. When attackers can run AI-driven attacks this cheaply and at this scale, defenders need AI on their side too.
For most security teams, that isn’t the reality yet, and their fix cycles are falling behind. Attackers now exploit vulnerabilities on average about seven days before a fix is available. AI is also growing the backlog from both ends: more code is written by AI, and more AI is scanning that code for flaws. Enterprises can’t fix every one of those vulnerabilities. With attackers moving this fast, security leaders keep asking the same questions: What should we fix first? Which exposures can an attacker use today? Which attacks do our existing mitigations already block?
Surgical validation, built on environmental context
Security tools have tried to answer these questions in two ways. Some reason about whether a vulnerability is exploitable without ever running an exploit. Others run exploits without the context of your full environment: which assets matter, which controls sit in the path, and who owns the fix. These same tools also promise to attack the whole production environment around the clock, which is risky, hard to scale, and expensive for a large organization. Teams are left with exploitable paths and no clear answer on how to stop them.
Zafran is the only platform that brings environmental context, proof, and action together, so teams can move from managing exposure to continuously preventing it.
That context makes offensive testing surgical. The Zafran Exposure Graph maps your assets, vulnerabilities, security controls, and the paths between them. It shows whether a vulnerable asset is running and reachable by an attacker, whether the vulnerability is being exploited in the wild, and whether your existing controls would block the attack. That narrows millions of findings down to the short list an attacker can actually reach and use, and the Offensive Validation agents test those.
The agents start from Zafran’s most valuable IP, the Exposure Graph, which is built for enterprise security teams and already knows your environment. Beyond known findings, they also look for unknown exposures, such as logic flaws and broken authorization in internet-facing apps and APIs. Because the testing is this focused, it is more efficient and scalable than attacking your whole environment, so teams can validate continuously. Every result feeds back into the Exposure Graph, so each proven exploit, blocked attack, or verified fix sharpens what Zafran tests and fixes next.
Zafran finds what’s exploitable, whether it’s a single flaw or a chain of low and medium severity vulnerabilities to your crown jewels. The Exposure Graph maps those chains, Offensive Validation proves the first step an attacker can exploit on each path, and Zafran mobilizes remediation. When a team can show a working exploit, the fix goes out right away. When a mitigation is available and blocks the attack, the team can schedule the patch for the next maintenance window with confidence.
How Offensive Validation works
The Offensive Validation agents work like a real attacker. They run recon, react to what they find, and adapt the attack as they go. They start from the assets already in the Exposure Graph, so there is no separate discovery tool to run. And because Zafran already runs on-prem, they can test internal assets and lateral movement paths along with internet-facing domains, web apps, and APIs.
Offensive Validation runs five types of validation. Each one ends in proof and, when needed, a clear next action:
- Validate unknown exposures. Agents test the internet-facing targets you choose, from whole domains down to a single API, for what scanners miss, such as logic flaws, broken authorization, and forgotten assets that never receive a CVE. That could be an admin endpoint that exposes live cloud credentials without a login, or an API that lets one customer read another customer’s data. A separate verifier proves the impact before anything is reported, and each proven issue becomes a new finding routed to the team that owns it.
- Validate exploitability of known findings. Zafran tests the findings it has already flagged as exploitable, on demand or on a schedule. Each test ends in one of three verdicts: proven exploitable, with evidence attached and the finding moved to the top of the queue; blocked by a control, which names the rule to keep in place; or not proven, which shows everything the agent tried and leaves the score unchanged, because one failed attempt doesn’t prove a finding is safe to leave open.
- Validate zero day exposure. When a new zero day is disclosed, the Exposure Graph already knows which assets run the affected software, whether they are reachable, and which controls sit in front of them. Testing starts automatically on just those assets, so within hours you know whether you’re exposed and who needs to act.
- Validate mitigations. Zafran recommends specific mitigations, such as moving a rule from detect to prevent. The agent runs the attack with the rule off, then runs it again with the rule on, and Zafran checks the control’s own logs to confirm the rule blocked it. One proof extends to every asset behind the same rule, and the test runs again whenever the configuration changes. It also confirms that a fix at an attack chain’s choke point actually breaks the chain.
- Validate fixes. A closed ticket doesn’t show that a fix actually took place. After a patch or rule change, the agent runs the same exploit against the same asset to confirm the attack path is broken. If the fix holds, the finding closes as verified, with evidence your auditors can use. If the exploit still works, for example on a server behind a load balancer that was missed, the finding reopens automatically with proof and goes back to the owner.
Your change management, enforced in code
The first question every security leader asks is whether this is safe to point at production. Because Zafran already knows each asset and the controls in front of it, the agents use only the techniques needed to prove impact.
You decide what the agent tests and how often it runs, and you can write your own rules of engagement in plain language. You also set how far it can go: recon only, safe checks, or supervised exploitation that asks for your approval before each step. Offensive Validation enforces those rules in its infrastructure. Each run happens in an isolated sandbox that is wiped afterward. An egress firewall limits the agent to the targets you approved, so if an asset is out of scope, the agent cannot reach it, even when the Exposure Graph knows it exists. The agent stops once it has proof, with no persistence, pivoting, or damage.
Test traffic comes from dedicated, published IPs that your SOC, incident response team, and MSSP can recognize. Every action is logged with a timestamp and source IP, so your team can replay what the agent knew, what it sent, what came back, and why it took the next step, then match it line by line to your own logs.
From proof to a verified fix
Offensive Validation is where insight turns into action that prevents exposure. Standalone validation tools can bury teams in findings with no workflow to fix them. Zafran pulls the raw configurations of the security tools in your environment, along with who owns each asset and each fix, so it can act on proven findings in two ways. Agentic remediation routes each fix to its owner in Jira or ServiceNow with the evidence attached. Live mitigation puts a surgical mitigation in place to break the exploitable path while the patch is on its way. Zafran re-tests every patch and mitigation so a finding only closes after a test confirms the fix.
Moving from exposure management to exposure prevention
When attackers move at machine speed, detection and response alone leaves little time to act. Prevention means closing the exposure before the attack starts. Zafran’s response pairs the context customers rely on today with offensive testing that confirms the exposures at the top of the list. Together, they make autonomous exposure prevention a continuous loop for the AI era: find the exposures and attack chains that matter, test them, fix them, and test again to confirm the fix holds. Fewer exposures, closed faster.