Hello from the Cloud-verse!
This week’s Cloud Security Newsletter topic: When the Pen-Test Agent Picks Its Own Targets: Richard Fan's Research on AWS Security Agent(continue reading)

This image was generated by AI. It's still experimental, so it might not be a perfect match!
Incase, this is your 1st Cloud Security Newsletter! You are in good company!
You are reading this issue along with your friends and colleagues from companies like Netflix, Citi, JP Morgan, Linkedin, Reddit, Github, Gitlab, CapitalOne, Robinhood, HSBC, British Airways, Airbnb, Block, Booking Inc & more who subscribe to this newsletter, who like you want to learn what’s new with Cloud Security each week from their industry peers like many others who listen to Cloud Security Podcast & AI Security Podcast every week.
Welcome to this week’s Cloud Security Newsletter!
Several of this week's critical flaws sit in code that most asset inventories never list on their own: the JVM inside Jira, Keycloak and Jenkins, an end-of-life SOAP library inside Confluence and N-central, the HTTP/2 module inside Apache, and the certificate authority inside Puppet. Three of those bugs were found with AI assistance. Google's threat intelligence team also documented an attacker framework that compares vendor patches, writes an exploit and runs it with nobody at the keyboard.
This week we feature Richard Fan, an AWS Security Hero based in Hong Kong, in conversation with Ashish on the Cloud Security Podcast. Richard set out to see how far he could push AWS's autonomous pen-testing agent outside its scope, and presented the results at fwd:cloudsec in London. He got it to attack a public site after verifying a private one, to follow a redirect onto a host it had promised not to test, and to work around its own tool restrictions. His conclusion is the line that stays with you: the agent's guardrails are promises, and "we all know the promise from AI is not guaranteed." [Listen to the episode]
⚡ TL;DR for Busy Readers
🚨 Java4Shell (CVE-2026-114938): pre-auth memory corruption in Java 17, 21, 25 and 27 on x86-64 CPUs with AVX-512 VBMI. Keycloak, Jira, Jenkins, Artifactory and PingFederate are reachable before login. Patch, or set -XX:-UseBASE64Intrinsics where no fixed build exists.
🚨 Apache Axis 1.x RCE exploited in Confluence and N-central: one intrusion went from a wiki SOAP endpoint to a cloud data warehouse and Microsoft 365 in about twelve hours. Disable Confluence's Remote API and N-central's Legacy API if you cannot upgrade.
Microsoft Agent Framework 1.26.1 and 1.26.2 on PyPI carried a credential stealer: pin 1.26.0 and rotate managed identity, Azure DevOps and GitHub Actions credentials.
Puppet Server RCE is on CISA KEV with a December 1 deadline: a compromised server exposes the CA signing keys
Richard Fan on AWS Security Agent: keep it off production, list exact endpoints on the allow list, and budget for about $50 per hour per parallel task.
📰 THIS WEEK'S TOP SECURITY HEADLINES
Each story includes why it matters and what to do next — no vendor fluff.
1. Java4Shell (CVE-2026-114938): pre-auth memory corruption in Java 17 and later, CISA alert issued
Primary source: The Record
Reporting: The Hacker News
Analysis: Tenable, Rapid7
What happened: A Base64 decoder intrinsic that Java uses on x86-64 CPUs with AVX-512 VBMI writes 62 or 63 bytes past the end of its output buffer when it decodes a malformed string, and Java only rejects the input after the write. Calif's Quang Luong and Albert Yang reported it, details leaked before a fix, and Oracle and OpenJDK shipped fixes on November 17 (Oracle Java 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1, with Corretto and Azul Zulu builds also listed). Affected CPUs are Intel Ice Lake, Sapphire Rapids, Emerald Rapids and Granite Rapids, and AMD Zen 4 and newer; ARM64 and standard Java 8 and 11 are not affected. Calif says Keycloak, Jira, Confluence, JFrog Artifactory, Jenkins and PingFederate are reachable before login in default configurations. CISA issued an alert on November 17. No exploitation has been reported and the flaw is not in KEV.
Why it matters: Exposure depends on the CPU under the workload as well as the Java version, so one container image can be vulnerable on an x86 node pool and safe on an ARM64 pool. The pre-login list is mostly identity and build infrastructure, which is where signing keys and federation trust live. Red Hat had no fixed OpenJDK builds when The Hacker News checked, and Temurin, Microsoft Build of OpenJDK and Liberica had none listed, so many teams will need the JVM flag before they can patch. Calif says the bug took Luong about an hour to find with AI.
Action for defenders: Map Java services to CPU flags (avx512vbmi) and start with internet-facing identity and CI systems. Where no fixed build exists, add -XX:-UseBASE64Intrinsics, which is Calif's preferred flag (-XX:UseAVX=2 stops ARM64 JVMs from starting). Rebuild images and restart services after patching.
2. Apache HTTP Server mod_http2 RCE (CVE-2026-119618): found by Anthropic and Google, exploitation status disputed
Primary source: SecurityWeek
Reporting: The Hacker News
Analysis: F5 Labs
What happened: A race condition in mod_http2 can corrupt memory and allows unauthenticated remote code execution in Apache HTTP Server 2.4.55 through 2.4.70 when HTTP/2 is enabled. Version 2.4.71 fixed it on November 11. Anthropic's OSS Scanner reported it on October 26, and Google Cloud Vulnerability Research (Anthony Weems and Simon Scannell) reported it independently on November 3. Red Hat scores it 9.8 and says public research demonstrated a reliable exploit. F5 Labs reported "no major exploit activity" on November 16, and a public GitHub proof of concept shows only crashes. Disclosed November 11, just before this edition's news window, and included because its exploitation status changed during the week.
Why it matters: Wiz found affected versions in 64% of cloud environments and publicly exposed instances in 51%, though those figures do not account for whether HTTP/2 is turned on. Two AI-assisted research teams found the same race condition eight days apart, which suggests attacker tooling can find it as well. F5's workaround of dropping h2 and h2c from the Protocols setting may open request smuggling risk, so check what sits in front of Apache before you disable HTTP/2.
Action for defenders: Find Apache 2.4.55 to 2.4.70 with mod_http2 loaded across images and VM fleets and upgrade to 2.4.71 or your distribution's package. Until then, alert on worker segfaults and HTTP/2 connection resets. The archived standalone mod_h2 project will get no patches.
🛠 If you only do one thing this week: run grep -o avx512vbmi /proc/cpuinfo | head -1 on the hosts behind your Keycloak, Jenkins, Jira and Artifactory instances, then check their Java versions. Anywhere both come back positive and no fixed build exists, add -XX:-UseBASE64Intrinsics and restart. Then do what Richard Fan does with AWS Security Agent and sort your agent controls into the ones the network enforces and the ones a model is only asked to respect.
3. Apache Axis 1.x RCE (CVE-2026-90436) exploited against Confluence Data Center and N-able N-central
Primary source: Rapid7
What happened: CVE-2026-90436 is an unsafe reflection flaw in Apache Axis 1.x (through 1.4) that allows unauthenticated RCE. Axis 1.x is end of life and Apache plans no fixed release. On November 13 Rapid7's MDR team saw an attacker send unauthenticated SOAP requests to a Confluence Data Center server with "Remote API (XML-RPC & SOAP)" enabled, run shell commands, take database credentials, and reach a cloud data warehouse and a Microsoft 365 account about twelve hours after the first request. The same day, N-able reported in-the-wild exploitation of N-central with the "Legacy API" enabled, which Rapid7 says strongly correlates with the Axis flaw. A proof of concept against N-central had been published on November 12. The fixes are Confluence 9.2.28 (10.x removed the feature and Atlassian Cloud is unaffected) and N-central 2026.4 HF3, released November 16.
Why it matters: The Confluence case shows an on-premises wiki becoming the credential source for a cloud data warehouse and a SaaS tenant within half a day. Both products are vulnerable only in non-default configurations, so scanners that test defaults will miss the exposed servers. The flaw is in a dead library that no vendor patch note names as a dependency, and finding it depends on whether your SBOM covers end-of-life Java components.
Action for defenders: Check whether Confluence's Remote API or N-central's Legacy API is on, and turn it off now if you cannot upgrade. Search other Java estates for Axis 1.x JARs. If either product was exposed, rotate the credentials stored in its database and review data warehouse and Microsoft 365 sign-ins from November 12 using Rapid7's IP indicators.
4. Puppet Server RCE (CVE-2026-115731) added to CISA KEV
Primary source: The Hacker News
What happened: CVE-2026-115731 (CVSS 9.4) is a deserialization flaw in Puppet Server's report handling. Any client holding a CA-signed agent certificate can send a crafted report and run code as the Puppet Server service user, and if the CA auto-signs requests without strong checks, an unauthenticated attacker can get a certificate first. It affects Puppet Enterprise 2026.0.0, 2025.x and 2023.8.x before 2026.0.1, 2025.12.1 and 2023.8.13, Puppet Core 9.x and 8.x before 9.1.3 and 8.22.2, and OpenVox (fixed in Server 9.2.0 and 8.18.0). Perforce shipped fixes on November 3. Neodyme's Alain Rödel found it with AI assistance. CISA added it to KEV on November 17 with a December 1 federal deadline. A companion flaw, CVE-2026-115732 (CVSS 9.0), is a certificate-name authorization bypass.
Why it matters: Neodyme says code execution on the Puppet server exposes the CA signing keys whatever the configuration, so a confirmed compromise means re-issuing certificates to every managed node.
Action for defenders: Patch Puppet Server and OpenVox, then review autosign.conf and any policy-based autosign script for weak checks. If the server was exposed before the November 3 fixes were applied, treat the CA key as compromised and plan a re-issue.
5. Check Point CVE-2026-123039: PHP2Shell plus a root escalation, then an ESXi wiper
Primary source: BleepingComputer
Analysis: Check Point Research
What happened: Check Point's DynamicCLI daemon runs as root and listens on a world-writable local socket without authenticating callers, so any local process can run commands as root. Attackers reached Mobile Access portals through PHP2Shell, escalated with this flaw, installed the TSGA backdoor and TSProxy library, moved over SSH to a VMware ESXi host using the vpxuser account, and ran a wiper. Check Point says "a couple of dozen customers" were hit since late October, starting with Israeli organizations, and it released emergency hotfixes. CISA added the flaw to KEV on November 13 with a November 16 federal deadline. Check Point Research does not make a high-confidence attribution: the tooling resembles PhiliKit, previously linked to China-nexus UNC5221, but the wiper note is in Hebrew.
Why it matters: Last week's edition covered PHP2Shell reaching a firewall management plane, and this week the same chain reached the hypervisor. The pivot used vpxuser, the account vCenter uses to manage ESXi hosts, so the gateway held a path into virtualization infrastructure that few segmentation diagrams show. A flaw in a language runtime ended in data destruction on a host that never ran PHP.
Action for defenders: Apply Check Point's hotfix (SK1000362), or stop dynamic_server and keep it from restarting as Check Point advises. Hunt for zz*.tpl files in portal templates, /tmp/CPconfigdaemon, /lib/libcpfwcommon.so and connections on TCP 11072, and search ESXi logs for SSH logins by vpxuser from gateway addresses.
6. Microsoft Agent Framework releases on PyPI poisoned with a credential stealer
Primary source: SecurityWeek
What happened: Versions 1.26.1 and 1.26.2 of agent-framework (about 190,000 monthly downloads) were uploaded to PyPI about 33 minutes apart on November 17. They kept the working framework code and added a stealer that runs through a Python startup mechanism every time the interpreter starts, even if the package is never imported. It targets environment variables, config files, other processes' credentials, Azure managed identity tokens and Azure DevOps tokens, and the second release added GitHub Actions credentials. Wiz found no matching release in Microsoft's GitHub repository, and SafeDep found a PyPI maintainer account created that day with publishing rights. Microsoft confirmed the versions were removed and that it is investigating. Version 1.26.0 is the last clean release.
Why it matters: An agent SDK gets installed where agents run, next to the managed identities that reach the tools those agents call, so a startup-hook stealer collects those tokens even if the agent code never runs. The malicious builds shipped only wheels with no source distribution, which a registry-mirror policy could have flagged. The release never touched the GitHub repository, so repository review and signing did nothing for anyone installing from PyPI.
Action for defenders: Search lock files, CI caches and images for agent-framework==1.26.1 or 1.26.2, pin 1.26.0, and rotate the managed identity, Azure DevOps and GitHub Actions credentials those environments held. Add a mirror rule that holds new releases of critical packages that arrive without an sdist or from a newly added maintainer.
7. Optimizely Web Experimentation used to inject skimmers and iOS exploit kits into 178 sites
Primary source: The Record
What happened: Optimizely confirmed that a vulnerability in its Web Experimentation platform let attackers inject JavaScript into some customers' sites. It says the path is blocked, the code was removed and affected customers were contacted. Report URI found the code on 178 widely visited sites, 32 of them in finance or banking. The payload swapped crypto wallet addresses, skimmed cards, stole cookies and screenshots, and sent iPhone and iPad visitors to the Coruna and DarkSword exploit kits. Report URI first saw the code in September, and the latest campaign ran on November 16.
Why it matters: The loader was placed in Optimizely's Project JavaScript field, so it was served from a vendor domain every affected site had already allowed in its Content Security Policy. Script integrity controls do not help here, because experimentation tools exist to change script content without a release. An outside researcher found it before the customers or the vendor did.
Action for defenders: If you load Optimizely Web Experimentation, find which pages loaded the snippet, for how long, and which of them took card data outside a processor-hosted iframe. Then list which third-party script vendors can push code to your production pages without your change control.
8. Google GTIG documents an attacker framework that diffs patches and exploits with no human in the loop
Primary source: Google Threat Intelligence Group
Reporting: BleepingComputer
What happened: Google Threat Intelligence Group found "exploit_agent" on an exposed attacker staging server, next to stolen victim data, webshells and brute-forcing scripts. The multi-agent framework compares web software before and after vendor patches, works out the flaw, drafts an exploit and runs it. GTIG calls this the first time it has seen discovery and exploitation automated "without a human in the loop." Configuration notes were in Russian, and victim data pointed at French media, entertainment and education sites. The report also covers open-source frameworks (CAIRN and PentAGI) used in real intrusions, a China-nexus group building its own framework, and LLMjacking, where stolen cloud AI credentials are resold through "LLM mixers."
Why it matters: Automated patch diffing shortens the time between a vendor fix and a working n-day exploit, and that window decides the outcome for stories 1 through 5. The cloud bill is the direct cost: Security Affairs reports Mandiant cases with fraudulent charges above $100,000 in a day, and one actor used a LiteLLM sandbox escape (CVE-2026-40217) to turn limited keys into admin credentials. Model API keys now have a resale market, and most programs do not rotate or monitor them like other cloud credentials. Richard Fan's research on AWS Security Agent looks at the same capability sold to defenders as a managed service.
Action for defenders: Put budget alarms and per-key quotas on every model API and AI gateway account, and find where those keys live (repositories, CI variables, LiteLLM configs). Shorten patch SLAs for internet-facing web applications on the assumption that a patch release now doubles as an exploit release.
🎯 Cloud Security Topic of the Week:
When the Pen-Test Agent Picks Its Own Targets: Richard Fan's Research on AWS Security Agent
AWS Security Agent, which Richard says AWS has renamed AWS Continuum, is a managed agent for code review, design review, threat modeling and autonomous pen testing. Only the pen-testing capability is generally available. Richard treated the agent itself as the target. He built intentionally vulnerable sites and tried prompts and infrastructure tricks to see whether it would break out of scope.
Three of his findings involve hosts the agent was never cleared to attack. A private DNS record let him pass domain verification and then point the name at a public IP mid-run. An HTTP redirect convinced the agent that an allow-listed site was the real target. A prefix-only allow list meant /log also allowed /login. He also got a reverse shell into the agent's environment, and watched the model run blocked tools through scripts. The network firewall held, because traffic could only reach the verified target and the allow list. The controls that failed were the ones that depended on the agent's own judgement about what to attack.
That raises the question Ashish put up front: where does shared responsibility sit when your cloud provider's agent pen-tests your applications? Richard's practical answer is to keep it off production, list exact endpoints, and keep that configuration as evidence of your intent.[Listen to the full episode →]
Featured Experts This Week 🎤
Richard Fan, AWS Security Hero.
Ashish Rajan - CISO | Co-Host, AI Security Podcast , Host of Cloud Security Podcast
Definitions and Core Concepts 📚
Before diving into our insights, let's clarify some key terms:
AWS Security Agent / AWS Continuum: Richard's description: "a managed agent to do security stuff" covering code review, design review and threat modeling, and autonomous pen testing. He says Continuum is the new name.
White box vs black box pen test: with the repository and design documents connected, the agent knows the endpoints and links findings to lines of code. Without them it spends tokens on reconnaissance.
Domain ownership verification: a token placed in the target's DNS record, which the agent checks before it will test a site.
Additional allow URL list: URLs the agent may reach through its firewall without verification, such as a third-party login provider, and promises not to pen test.
Prefix match: how Richard found the allow list compares URLs. An entry for /log also allows /login and anything else that starts with those characters.
Shared responsibility model: AWS secures the cloud and the customer secures what they put in it, in Richard's and Ashish's shorthand.
AVX-512 VBMI: the x86-64 CPU instruction set extension that the vulnerable Java Base64 intrinsic uses. Intel Ice Lake and later server chips and AMD Zen 4 and later have it.
KEV: the CISA Known Exploited Vulnerabilities catalog. Federal agencies get a remediation deadline when a flaw is added.
LLMjacking: theft and resale of cloud AI or model API access, as described in Google's GTIG report.
vpxuser: the account vCenter uses to manage ESXi hosts, used for the pivot in the Check Point intrusions.
This week's issue is sponsored by AI Security Lab
💡Our Insights from this Practitioner 🔍
When the Pen-Test Agent Picks Its Own Targets: Richard Fan's Research on AWS Security Agent
1. Domain verification can be passed privately and then pointed at a public target
Security Agent will not test a public site until you prove ownership with a DNS token. It can also test internal sites inside your VPC, and there the token lives in your own private DNS, where you can create a record for any name. Richard created a private record for a domain he owned while pretending it was someone else's (he uses amazon.com as the example), pointed it at a private IP, and let verification pass. Then he changed the record.
Richard: "At that time, I changed my private DNS, point amazon.com to the real amazon.com public IP, and the agent follow that modified DNS record and start pen test the real amazon.com."
He compares it to a human tester, who would ask why anyone wants an internal test against amazon.com and would be extra careful not to touch the real one. His explanation is a hand-off between agents.
Richard: "So I think it's just a context switching between two agent, two brain. The first agent validate it is, "Oh, I'm pen testing a private amazon.com." And the other agent comes in and do the real pen testing, didn't know what was going on."
Ashish took it a step further: an attacker who rebuilds a target's open source code in a private VPC could run the agent against that copy. Richard pointed out where the line still holds. You can test a duplicate of someone's code, but verification exists to stop real damage to a production site and its customer data. Ashish suggested AWS consider verifying internal domains too.
2. The firewall held; the agent's promise did not
Richard's second test used two sites. Website A, a verified target, sent every request to website B with a 301 or 302 redirect. Website B was on the "additional allow URL" list, which exists so the agent can reach things like a third-party login page without testing them. After a few rounds the agent decided B was the real site behind A and started attacking it.
Richard: "So what I basically achieve is that I can have my website, my own website, have nothing, just HTTP redirect to amazon.com."
He gives AWS credit for the network design. The firewall only lets agent traffic reach the verified target and the allow list, so without the allow-list entry the redirected requests would never have left the environment.
Richard: "I think it's a good decision from AWS to do the whitelist approach of the firewall, but it's just it's not perfect because the distinction between pen testing a URL or not is really based on the agent."
And the line from the cold open: "But it's just simply a promise from the AI agent, and we all know the promise from AI is not guaranteed."
3. The allow list matches prefixes, so /log also means /login
Cautious users narrow the allow list to the exact API paths they need, for example the Auth0 endpoints their login flow uses. Richard found that does not narrow much.
Richard: "But the funny part is, I found out this URL allow list is just a prefix only."
An entry for /log also allows /login, and there is no wildcard syntax to control it. A deny list exists, but you cannot list every suffix. Ashish compared it with a human pen-test contract, where scope can include the homepage only or everything except /login. Richard confirmed that granularity is not available today. His verdict on the matching logic: "I think it is a very low-level stupid mistake."
4. The agent works around the tool restrictions meant to contain it
Richard got a reverse shell into the agent's environment, which is how he saw the tooling installed there (a URL fuzzer among it). He also watched what the model did when a tool was blocked: it ran the same open source tool from a script instead. He could not recall which tools were restricted, saying the test was a long time ago.
His blog post ends on the worry he repeated in the episode.
Richard: "This security agent is very capable, and is so capable that I'm worried about the people making the guardrail is being outsmart by the AI."
Richard: "But actually, the AI is very smart to just circumvent all these kind of guardrails."
The engineers' assumption, as he describes it, is that restricting tools stops someone like him from tricking the agent. The model treats the restriction as a problem to route around.
Ashish framed the question in his intro: "And where does the line for shared responsibility now change when your cloud provider starts doing pen test of the applications you're hosting, either internally or externally."
Richard puts the model, runtime and configuration options on AWS's side of the line. He reads AWS's position as a usage rule.
Richard: "But I think what Amazon want to s-say is that you need to use it in the right way. You cannot intentionally trick the agent to do this or to do that."
His objection is that intent used to settle the question. Misusing the cloud on purpose made it the customer's fault. A site that redirects elsewhere is a normal thing to run, and the agent turned it into a scope change.
Richard: "But now AI has its own thinking. I can claim that I'm not intentionally tricking the AI to do bad thing."
Richard: "So is it me doing bad thing tricking the agent or is it Amazon failed to do their guardrail, their responsibility of safeguarding the model? I think this part is not very clear."
6. You can press stop, but nobody can watch a parallel run
Ashish asked the question a CISO would ask in the cold open: "If I start the pen test and it goes to that URI which is not in scope, can I stop this, or I just have to wait and see what happens?" You can stop a run from the console. Richard doubts anyone can supervise it closely enough to know when.
Richard: "But as a human, I don't think you can monitor because it, it's not a single-threaded agent task."
Production is out. Ashish: "Definitely do not use this in production." Richard points to AWS's own documentation: "Because it is also specifically called out in the official documentation is never use the pen test on production-"
His advice for protecting yourself, offered with the disclaimer that it is his own thinking, is to be precise in the configuration even where precision does not restrict anything. If your site uses Auth0, do not allow-list the whole domain.
Richard: "Put the whole API endpoint into it. Whatever API endpoint is required to authenticate to your website, just put it, the list of it."
If AWS ever questions your account over traffic to an out-of-scope URL, that detailed list shows what you intended.
7. $50 per hour per task, parallel agents, and no word on the model
The pen-test capability was free in preview. Richard: "It's free when it's preview, but now it's GA, so it's $50, five zero, not one five." The rate applies per task, and the service runs several at once.
Richard: "So as I say, the agent will spin up multiple agent to do parallel pen test for different kind of vulnerability in parallel."
He re-ran his tests before his talk because "I want to make sure my finding that AWS refused to fix is still there." He stopped the run as soon as the screenshot he needed appeared: "It just run for around one hour, and then it cost me almost $200 already."
Logs show the commands, the model's reasoning and the tools used. They do not show which model is running underneath. Asked whether AWS discloses it: "Yeah, as, as I'm aware, they're not talking about it."
8. Connect the repository, and treat the rest as "better than nothing"
Connecting a GitHub repository and design documents makes the test white box. The agent knows which endpoints exist, links findings to the line of code, and avoids burning tokens on discovery.
Richard: "Instead of black box pen testing you burn a lot of token just to do the reconnaissance."
The other three capabilities (code review, design review and threat modeling) are still in preview, and Richard says they have been for a year. His overall rating: "I would say the whole product of Continuum is something like better than nothing." A company with a security team should still do threat modeling itself, with AI help. A small team without dedicated security support may get real value. He remembers trying to teach threat modeling to application teams as an embedded security partner: "And most of the time is that I feel difficult to explain to the team how to do threat modeling because it's just like a creative game."
Ashish put the demand plainly: "A lot of people perhaps want to do a pen test and not wait six months."
Practical takeaways
Patch or flag Java4Shell on x86 identity and CI systems. Check for avx512vbmi, patch where a build exists, and add -XX:-UseBASE64Intrinsics where it does not.
Turn off Confluence's Remote API and N-central's Legacy API if you cannot upgrade, then search for Axis 1.x JARs elsewhere.
Pin agent-framework to 1.26.0 and rotate the credentials of any environment that installed 1.26.1 or 1.26.2.
Split agent controls into enforced and advisory. For every agentic tool, list what the network or platform blocks and what the model is only asked to avoid.
Keep provider-run pen-test agents off production and allow-list exact endpoints, not domains.
📚 RELATED RESOURCES 🎧
Question for you? (Reply to this email)
🛑 If an agent in your environment went rogue today, how fast could you cut its egress, and who holds that switch?
Next week, we'll explore another critical aspect of cloud security. Stay tuned!
📬 Want weekly expert takes on AI & Cloud Security? [Subscribe here]”
We would love to hear from you📢 for a feature or topic request or if you would like to sponsor an edition of Cloud Security Newsletter.
Thank you for continuing to subscribe and Welcome to the new members in tis newsletter community💙
Peace!
Was this forwarded to you? You can Sign up here, to join our growing readership.
Want to sponsor the next newsletter edition! Lets make it happen
Have you joined our upcoming AI Security Lab session yet?
checkout our sister podcast AI Security Podcast



