# What changed in the last 90 days

News, read 10 Nov 2026

**The short answer.** 17 dated items from 2 Sep to 18 Nov 2026 cover launches, research and acquisitions among the profiled platforms, such as Cogent Security acquiring Specular on 10 Nov 2026. Each item links the original announcement, and items are listed newest first by date, not by importance.

## Maze releases a free read only checker for the Java4Shell JDK flaw

18 Nov 2026. Maze published java4shell-check on 18 Nov 2026, a free open source, read only tool that assesses whether CVE-2026-114938 is exploitable on the JVMs an organization runs. The flaw is a heap out of bounds write in OpenJDK's AVX-512 Base64 decoder, with fixes released 17 Nov 2026.

Maze's write-up says exposure needs three conditions together: an affected JDK (for example 17.0.6 through 17.0.21, 21 through 21.0.13, 25 through 25.0.5 and 27 through 27.0.1 and builds derived from them), an x86-64 CPU with AVX-512 VBMI, and the Base64 intrinsic enabled, which it is by default. Maze says ARM64, OpenJ9 and standard Java 8 and 11 builds do not run the vulnerable code.

The tool runs where the JVMs run, sends nothing to other services and changes nothing. Maze suggests handing the repository to a coding agent with read only credentials, and states that a patched JDK on disk does not fix a JVM that is already running, so each JVM must be restarted after patching.

Maze reports a CVSS 4.0 rating of 9.2, credits the finder as Calif, and says no exploitation in the wild had been reported as of 18 Nov 2026. The OpenJDK advisory of 17 Nov 2026 lists the flaw as CVE-2026-114938 in hotspot/compiler. Product claims about the checker are Maze's.

Source: [mazehq.com](https://mazehq.com/blog/java4shell), read 19 Nov 2026.

## Cognizant joins the Qualys mROC Alliance for managed remediation services

18 Nov 2026. Qualys announced on 18 Nov 2026 that Cognizant has joined its Global mROC Alliance and will pair Cyber Vulnerability Management services with Qualys ETM and TruRisk Eliminate. Qualys says Cognizant will package four mROC services built on ETM.

The release, datelined Foster City, describes a three step operating model: detection with Qualys InstaScan, prioritization with Qualys Agent Val, which validates which vulnerabilities are exploitable in a client environment, and autonomous remediation with TruRisk Eliminate with TruConfidence. Qualys says the remediation step accounts for existing security controls, selects a patch, mitigation or isolation, verifies the fix and rolls back if system health declines.

Qualys states that on November's Patch Tuesday InstaScan delivered 90 percent of detections within minutes of disclosure, and that of 48,172 vulnerabilities disclosed in 2025 its Threat Research Unit found 357 (0.74 percent) confirmed remotely exploitable and actively weaponized. These are vendor figures.

Cognizant will work within each client's existing tools, change windows and approval processes, agree scope with application owners, and phase in automation starting with low risk systems. The four services span cyber risk quantification and advisory, onboarding and integration, managed risk monitoring and managed risk remediation. Vishal Salvi of Cognizant is quoted in the release.

Source: [www.qualys.com](https://www.qualys.com/company/newsroom/news-releases/usa/cognizant-joins-qualys-global-mroc-alliance-deliver-managed-risk-operations-frontier-ai-era), read 19 Nov 2026.

## Zafran announces Offensive Validation for its exposure platform

18 Nov 2026. Zafran announced Offensive Validation on 18 Nov 2026 as a new part of its Autonomous Exposure Prevention platform. AI agents test an environment using the context in Zafran's Exposure Graph, to confirm which known findings an attacker can use and to look for unknown exposures.

Zafran says the feature completes a cycle of assess, validate, fix and verify. Each proven exposure is routed to the relevant owner with a fix that can be tested and verified in one click, and every result feeds back into the Exposure Graph.

The announcement says testing is narrowed by the graph, which maps assets, vulnerabilities, security controls and the paths between them, so agents test the short list an attacker could reach rather than the whole production environment. Beyond known findings, Zafran says the agents look for logic flaws and broken authorization in internet facing apps and APIs.

The post cites an FBI FLASH alert of 13 Nov 2026 about ARTEX, an open source AI agent penetration testing framework that it says attackers have used since at least August. That description is Zafran's account; the FBI alert itself was not read for this item. Claims about the platform are Zafran's own.

Source: [www.zafran.io](https://www.zafran.io/resources/zafran-announces-offensive-validation-to-prevent-exposure-ahead-of-ai-attackers), read 19 Nov 2026.

## Cogent introduces DefenseOS with Agent Policies and Agent Automations

17 Nov 2026. Cogent published on 17 Nov 2026 a description of DefenseOS, the context engine and agent runtime inside its platform, and introduced two capabilities: Agent Policies and Agent Automations. Cogent states DefenseOS ingests data through more than 100 integrations and that Agent Automations ships with 30 or more templates.

The post, by Josh Dreyfuss, says DefenseOS has two jobs. The context layer ingests scanner, business-system and security-tool data, normalizes and deduplicates findings, and links each vulnerability to the assets, owners and dependencies it touches. The execution layer is the harness the agents run inside, with approval gates, confidence thresholds and policy limits that decide what an agent may do alone. Cogent says every action is logged and traces back to its recommendation, and that low-confidence calls go to a person.

Agent Policies turns what DefenseOS learns into rules a team can read, edit and approve. Cogent lists three modes: rule discovery (agents propose, people accept, edit or reject), rule creation from uploaded documents or natural language, and drift detection that proposes updates when context changes.

Agent Automations runs recurring work on a schedule or on events, from a library of 30 or more templates or built in natural language, with results sent to Slack or email. These figures and the single unnamed customer example (nine reporting automations) are Cogent's own statements; the post gives no independent measurement. Cogent says most teams start with agents recommending fixes that people approve, then widen autonomy.

Source: [www.cogent.com](https://www.cogent.com/blog/introducing-defenseos-the-operating-system-that-runs-your-cyber-defense-factory), read 19 Nov 2026.

## Rival publishes a study of a model safety classifier that blocks security agents

12 Nov 2026. Rival Security published research on 12 Nov 2026 on how often a frontier model's cyber safety classifier blocks its vulnerability analysis agents. It reports blocked threads rising from about 5 percent to about 20 percent on some days in early November 2026.

Rival says its agents download vulnerable open source packages in a sandbox, read the code and decide when a flaw is exploitable, and that they run on gpt-6.1-sol. The study sampled 2,000 blocked threads from 14 days of production traces and replayed blocked requests against the real model and classifier, changing one thing at a time.

Rival reports that network facing packages were blocked 16.1 percent of the time against 6.2 percent for other packages, a ratio of 2.6, and that 33 of 78 blocked requests were never blocked again when resent unchanged. These are Rival's findings from its own traces.

The post notes that a block makes the run fail or fall back to a weaker model, which matters to anyone building security agents on these models. It does not publish exploit material.

Source: [rival.security](https://rival.security/blog/openai-cyber-classifier-security-agents?utm_source=automatedvulnerabilityremediation.com&utm_medium=referral&utm_campaign=directory), read 19 Nov 2026.

## Cogent Security acquires Specular

10 Nov 2026. Cogent Security announced on 10 Nov 2026 that it had acquired Specular, a Google-backed AI security platform, and that Specular founder Peyton Smith and team members are joining Cogent. The company says the deal strengthens its ability to find, validate and fix vulnerabilities, and ties it to its PHP2Shell analysis.

Cogent's blog post, dated 10 Nov 2026, says Specular helps enterprises see their environments the way sophisticated attackers do.

The post cites Cogent's PHP2Shell analysis: when the flaw was disclosed on 21 Oct 2026, the average Fortune 500 company in its research had 2,266 affected assets, and Cogent says it narrowed that to 16 using runtime exposure, internet reachability and business criticality.

These figures are Cogent's own claims.

Source: [www.cogent.com](https://www.cogent.com/blog/cogent-expands-its-ai-cyber-defense-factory-with-specular-acquisition), read 10 Nov 2026.

## Cogent reports VR-1.1 repair results on LiveCVEBench

3 Nov 2026. SiliconANGLE reported on 3 Nov 2026 that Cogent Security's VR-1.1 model produced a working repair for 57.9% of the 190 tasks in LiveCVEBench, while Claude Code and Codex each scored 45.3%. Cogent ran the test. The report notes it blocked outbound network access during repairs.

The report says grading tests were hidden until each agent finished, and that inference budgets differed across the systems compared.

SiliconANGLE also reports that each patch goes to a separate review that starts from a fresh context and looks for a way around the fix, and that the model leaves regression tests.

The result is a vendor-run benchmark reported by a news outlet, and the register labels it that way.

Source: [siliconangle.com](https://siliconangle.com/2026/11/03/exclusive-cogent-securitys-vr-1-1-model-outscores-claude-code-and-codex-on-vulnerability-repair/), read 10 Nov 2026.

## Averlon publishes a response write-up on a Linux kernel TCP flaw

2 Nov 2026. Averlon published a write-up on 2 Nov 2026 on CVE-2026-98432, a Linux kernel TCP use-after-free rated CVSS 9.8, which the kernel CNA published on 15 Oct 2026. The post is dated 18 days after the CVE and covers which network paths expose a host.

Averlon's post describes the flaw as a race condition in the kernel's IPv4 TCP handshake code, reachable by a remote attacker with no credentials.

The write-up arrives 18 days after the CVE; Averlon states no time-to-answer figure for new CVEs on the pages read for the register, which records none found on 10 Nov 2026.

Source: [www.averlon.ai](https://www.averlon.ai/blog/cve-2026-98432-linux-kernel-tcp-use-after-free), read 10 Nov 2026.

## Rapid7 covers further exploited Ivanti Connect Secure flaws

2 Nov 2026. Rapid7 published on 2 Nov 2026 that Ivanti updated its advisory on 29 Oct 2026 to add CVE-2026-111357 and CVE-2026-118527 and released fixes. CISA added both to its KEV catalog on 30 Oct 2026. Rapid7 says authenticated checks have been available since the 30 Oct content release.

Ivanti says exploitation in the wild occurred before disclosure, affecting a limited number of customers when the flaws were chained with CVE-2026-104935.

Rapid7 notes a scoring difference: it rates CVE-2026-111357 at 7.0 while Ivanti assesses 8.8, based on different attack complexity and scope.

The blog covers mitigation guidance and how Rapid7 customers can check exposure.

Source: [www.rapid7.com](https://www.rapid7.com/blog/post/etr-additional-ivanti-connect-secure-zero-days-exploited-in-the-wild-cve-2026-111357-cve-2026-118527/), read 10 Nov 2026.

## Rival publishes its PHP2Shell triage write-up

2 Nov 2026. Rival Security published a write-up on 2 Nov 2026 on how its agents triaged PHP2Shell, CVE-2026-109618 and CVE-2026-109619. It reports scanning more than 80,000 repositories within hours, and one unnamed customer where an agent found 60 PHP repositories, 29 potentially vulnerable.

The post says its agents began triaging customer environments hours before vulnerability scanners could detect the CVEs.

The post also says agents can open pull requests, suggest web application firewall rules, apply configurations and message the relevant employees.

Customer results in the post are unnamed, and the figures are Rival's own claims.

Source: [rival.security](https://rival.security/blog/php2shell-black-october-2026?utm_source=automatedvulnerabilityremediation.com&utm_medium=referral&utm_campaign=directory), read 10 Nov 2026.

## Onit announces its Zero-Day Agent ahead of launch

23 Oct 2026. Onit Security published a post on 23 Oct 2026 announcing its Zero-Day Agent, which it says catches a zero-day as it breaks, finds every exposed asset and shows which are exploitable, in minutes. The post says PHP2Shell leaked on 20 Oct 2026 and was patched about 18 hours later.

The post states the agent runs no live exploits against production, needs no new endpoint agent and takes no disruptive action without approval.

It says full launch comes the following week. The minutes claim carries no published method, and the register labels it the vendor's claim.

Source: [onit.security](https://onit.security/blog/zero-day-response/), read 10 Nov 2026.

## Qualys offers free PHP2Shell detection for 15 days

23 Oct 2026. Qualys announced on 23 Oct 2026 that it was offering its solutions free to the community to detect, validate and mitigate exposure to PHP2Shell, CVE-2026-109618 and CVE-2026-109619. The service is free for 15 days for up to 1,000 assets from activation, and attack traffic had already been reported.

The announcement describes both flaws as unauthenticated remote code execution, each rated CVSS v4.0 9.2.

The offer is a time-limited response to a single disclosure. Qualys states no result figures for it in the announcement.

Qualys Threat Research Unit data in the release covers 167 million PHP package installations, 73% of them on internet-facing assets. Of 156 million with a known version, 110 million (71%) run end-of-life or end-of-support versions, 5.4 to 8.1, and most of these will not get a vendor patch. Another 44 million (28%) run supported branches, PHP 8.2 to 8.5, where a fix exists. Only 581,000 installations (0.37%) were patched in the first 24 hours after disclosure.

The trial covers Qualys VMDR, Enterprise TruRisk Management and TruRisk Eliminate. The release lists five steps: discover, prioritize and validate, patch supported versions and restart PHP services, mitigate unsupported versions with OS vendor backports or isolation, then rescan to verify. It also says leaked PHP worker memory can expose passwords, tokens and API keys.

Source: [www.qualys.com](https://www.qualys.com/company/newsroom/news-releases/usa/media-alert-qualys-offers-free-services-for-php2shell), read 10 Nov 2026.

## Maze publishes its PHP2Shell write-up one day after the patch

22 Oct 2026. Maze Security published a write-up on 22 Oct 2026, a day after the 21 Oct 2026 patch release for PHP2Shell, CVE-2026-109618 and CVE-2026-109619. It states how to tell whether a server is affected, including PHP behind Apache with mod_php, where an unauthenticated attacker may run code.

The post says both flaws are rated CVSS 9.2 on v4.0 and were patched a day earlier than planned.

Maze's pages also state that agents sweep every instance for a new zero-day in hours, before scanners have a signature; that is the vendor's claim.

Source: [mazehq.com](https://mazehq.com/blog/php2shell), read 10 Nov 2026.

## Rapid7 puts average time to exploitation at three days

21 Oct 2026. Rapid7's Q3 2026 threat update, dated 21 Oct 2026, reports three days as the average time from public disclosure to exploitation of high-severity vulnerabilities, with 71% of tracked exploitation inside seven days. It tracked 43 newly exploited high and critical vulnerabilities in the quarter.

The report says exploitation occurred within 14 days for 80% and within 30 days for 90% of the tracked set.

Rapid7 says the 43 newly exploited vulnerabilities were down 7% from Q2 and up 19% on the same period a year earlier.

Of the 43 newly exploited vulnerabilities, 34 (79%) required no authentication or user interaction. Rapid7 also says 8% of the threat reporting it tracked in Q3 cited AI as an operational component, with human operators still selecting targets and objectives.

Underground forum monitoring in the quarter found 41 posts advertising RDP or network access. The post dated 21 Oct 2026 says the full Quarterly Threat Landscape Report also covers ransomware activity and APT tracking, and recommends starting with externally accessible assets and the vulnerabilities that affect them.

Source: [www.rapid7.com](https://www.rapid7.com/blog/post/tr-q3-2026-threat-update-attackers-exploiting-vulnerabilities-within-3-days/), read 10 Nov 2026.

## Tenable adds an AI Inspector to the CyberAgents Exchange

8 Oct 2026. Tenable announced on 8 Oct 2026 the CyberAgents Exchange AI Inspector, built with OpenAI GPT cyber models, to evaluate AI agents, skills and MCP servers listed on the Exchange. The Exchange is an open-source, vendor-agnostic hub of agents and playbooks for security teams.

Tenable says the Inspector includes an enterprise-grade safety review designed to help organizations deploy listed components more securely in production.

OpenAI's head of global cyber partnerships is quoted in the release on the trust layer needed to scale agentic AI.

Tenable describes Exchange Inspector as a three-part vetting process. It starts with the skills inspection engine in Tenable One AI Exposure, which is designed to detect malicious content such as prompt injection and exposed hardcoded secrets or customer data. The component then gets an assessment using OpenAI GPT cyber models, followed by review from Tenable's security research team.

The release names three vetted skills available on the date of the announcement: SOC Hunter, Remediation Priority & Impact Agent, and Splunk Tenable Cloud Security Skill. Tenable says Exchange Inspector was developed as part of its participation in the OpenAI Daybreak Defense Network. The hunt-time figure for SOC Hunter is reported by the contributor, Tenable's enterprise security team.

Source: [www.tenable.com](https://www.tenable.com/press-releases/tenable-uses-openai-gpt-cyber-models-to-advance-agentic-security-review-in-the-cyberagents-exchange), read 10 Nov 2026.

## Zafran reports Gartner Emerging Tech mentions

22 Sep 2026. Zafran announced on 22 Sep 2026 that it was named a Sample Vendor for autonomous exposure remediation in Gartner's Emerging Tech Impact Radar: Preemptive Cybersecurity, published 11 Sep 2026. It also cites an April 2026 Gartner report on funded startups, covering 148 startups and about $4.19 billion.

The announcement repeats Zafran's own analysis that about 90% of vulnerabilities can be mitigated with controls already in place, such as firewalls, endpoint protection and web application firewalls.

The recognition is reported by Zafran; the register records it as the vendor's claim.

On the page, Zafran says it was named a Sample Vendor for Autonomous Exposure Remediation in the Gartner report. A Sample Vendor entry is an example named in a technology category; the page quotes Gartner's definition of the category as generative AI orchestrating closed-loop remediation that generates, validates and deploys fixes across code, supply chain and infrastructure.

Zafran also says it was identified among funded startups in the Unified Exposure Management Platform category in a Gartner report published 3 April 2026. The page quotes Gartner as saying that by 2028 at least half of the exposure management market will consist of unified exposure management platforms, up from less than 5% in 2025.

Source: [www.zafran.io](https://www.zafran.io/resources/zafran-recognized-in-two-2026-gartner-r-emerging-tech-reports-for-unified-exposure-management-platforms-and-autonomous-exposure-remediation-categories), read 10 Nov 2026.

## Seemplicity releases Response Options

2 Sep 2026. Seemplicity announced Response Options on 2 Sep 2026, available to all customers. They give alternatives to a full patch, such as compensating controls and targeted configuration changes, each with a Readiness Indicator showing whether it is ready to deploy now or needs review first.

AI Analysts pre-select a recommended option, and every alternative stays visible with its reasoning.

Options appear in the Analyst Results tab on a finding's detail page. The release states no customer results.

The release, datelined Las Vegas, describes the problem as follows: a complete fix such as a patch, an upgrade or a major configuration change is seen as the safest long-term answer, but it is also the slowest and most intrusive. It can demand heavy testing, coordination and sometimes a production reboot.

Seemplicity says Response Options is meant to give faster, lower-risk steps that reduce exposure now, without waiting on the full remediation cycle. Ravid Circus, Chief Product Officer and Co-Founder, is quoted in the release saying that patch or nothing is what slows organizations down. The capability applies to vulnerability management and exposure management workflows.

Source: [seemplicity.ai](https://seemplicity.ai/news/response-options/), read 10 Nov 2026.

## Related pages

- [Remediation Register: the data, downloadable](/datasets/)
- [How the register works, and what it leaves out](/method/)
- [The Backlog Calculator](/calculator/)
- [Guides](/guides/)

## Sources

- https://mazehq.com/blog/java4shell (read 19 Nov 2026)
- https://www.qualys.com/company/newsroom/news-releases/usa/cognizant-joins-qualys-global-mroc-alliance-deliver-managed-risk-operations-frontier-ai-era (read 19 Nov 2026)
- https://www.zafran.io/resources/zafran-announces-offensive-validation-to-prevent-exposure-ahead-of-ai-attackers (read 19 Nov 2026)
- https://www.cogent.com/blog/introducing-defenseos-the-operating-system-that-runs-your-cyber-defense-factory (read 19 Nov 2026)
- https://rival.security/blog/openai-cyber-classifier-security-agents (read 19 Nov 2026)
- https://www.cogent.com/blog/cogent-expands-its-ai-cyber-defense-factory-with-specular-acquisition (read 10 Nov 2026)
- https://siliconangle.com/2026/11/03/exclusive-cogent-securitys-vr-1-1-model-outscores-claude-code-and-codex-on-vulnerability-repair/ (read 10 Nov 2026)
- https://www.averlon.ai/blog/cve-2026-98432-linux-kernel-tcp-use-after-free (read 10 Nov 2026)
- https://www.rapid7.com/blog/post/etr-additional-ivanti-connect-secure-zero-days-exploited-in-the-wild-cve-2026-111357-cve-2026-118527/ (read 10 Nov 2026)
- https://rival.security/blog/php2shell-black-october-2026 (read 10 Nov 2026)
- https://onit.security/blog/zero-day-response/ (read 10 Nov 2026)
- https://www.qualys.com/company/newsroom/news-releases/usa/media-alert-qualys-offers-free-services-for-php2shell (read 10 Nov 2026)
- https://mazehq.com/blog/php2shell (read 10 Nov 2026)
- https://www.rapid7.com/blog/post/tr-q3-2026-threat-update-attackers-exploiting-vulnerabilities-within-3-days/ (read 10 Nov 2026)
- https://www.tenable.com/press-releases/tenable-uses-openai-gpt-cyber-models-to-advance-agentic-security-review-in-the-cyberagents-exchange (read 10 Nov 2026)
- https://www.zafran.io/resources/zafran-recognized-in-two-2026-gartner-r-emerging-tech-reports-for-unified-exposure-management-platforms-and-autonomous-exposure-remediation-categories (read 10 Nov 2026)
- https://seemplicity.ai/news/response-options/ (read 10 Nov 2026)

Last reviewed 10 Nov 2026. Canonical: https://automatedvulnerabilityremediation.com/news/
