SEC595: Applied Data Science and AI/Machine Learning for Cybersecurity Professionals
SEC595Cyber Defense, Artificial Intelligence
SANS Community Benefits
Connect, learn, and share with other cybersecurity professionals
Top of the News
Patch Tuesday: Microsoft, Adobe, SAP
On Tuesday, April 14, Microsoft released updates to address 167 vulnerabilities in their products, as well as dozens of vulnerabilities in Chromium. One of the Microsoft vulnerabilities, a spoofing vulnerability in Microsoft Office SharePoint (CVE-2026-32201) is being actively exploited, and another, a privilege elevation vulnerability in Microsoft Defender (CVE-2026-33825) was previously disclosed. Adobe released updates to address 55 vulnerabilities in 11 products, including five critical vulnerabilities in Adobe Cold Fusion: two security feature bypass flaws, two arbitrary code execution flaws, and an arbitrary file system read flaw. Adobe also released fixes for critical vulnerabilities in Acrobat Reader, InDesign, InCopy, FrameMaker, Connect, Bridge, Photoshop, and Illustrator. Adobe also addressed several other important vulnerabilities, including arbitrary code execution flaws, application denial-of-service flaws, and a memory exposure issue, in Adobe Experience Manager (AEM) and Adobe DNG Software Development Kit (SDK). SAP released 22 new and updated security notes, including addressing a critical SQL injection vulnerability in Business Planning and Consolidation and Business Warehouse (CVE-2026-27681).
Editor's Notes
- Slide 1 of 3
That’s a BIG number of flaws there, and I think they are going to get bigger and bigger in coming months. It’s time to rev up your patching processes to deal with the new reality. Some folks are starting to refer to "VulnOps," a concept whereby you have dedicated staff and autonomous (and perhaps AI-driven) discovery/patching of vulnerabilities in your environment. I think there needs to be a little more fleshing out of the concept of VulnOps at this point, but it’s a concept worth keeping an eye on.
- Slide 2 of 3
Adobe, Google, Microsoft, and Fortinet already have your attention; don’t overlook those SAP flaws. These are critical as well, and you’re going to face a more robust change control than you did with the other updates. Maybe a good time to see just how automated the processes are for the first three vendors to see if you can improve those processes further to free up resources for more involved updates like SAP.
- Slide 3 of 3
Largest number in a while. Mythos?
Slide 1 of 0- Slide 1 of 3
Fortinet Patches Two Critical Flaws in FortiSandbox
Fortinet's Product Security Incident Response Team (PSIRT) issued patches for more than 25 vulnerabilities on Tuesday, April 14, 2026, including two critical-severity and three high-severity flaws. CVE-2026-39808 allows an unauthenticated attacker to execute arbitrary code or commands using crafted HTTP requests, due to an OS command injection vulnerability in Fortinet FortiSandbox 4.4.0 through 4.4.8. CVE-2026-39813 also affects FortiSandbox 4.4.0 through 4.4.8 as well as 5.0.0 through 5.0.5, and allows an unauthenticated attacker to bypass authentication using crafted HTTP requests, due to a path traversal vulnerability in the JRPC API. Both flaws carry CVSS score 9.1; users should update to fixed versions of FortiSandbox 4.4.9 or later, or 5.0.6 or later. Two of the high-severity flaws are SQL injection vulnerabilities, in FortiDDoS-F and FortiClientEMS respectively, allowing authenticated arbitrary SQL queries (CVE-2026-39815, CVSS 7.9; CVE-2026-39809, CVSS 7.1), and the third is a heap-based buffer overflow vulnerability in the FortiAnalyzer Cloud oftpd daemon allowing unauthenticated remote code execution (CVE-2026-22828, CVSS 7.3).
Editor's Notes
- Slide 1 of 3
Wow! More than 25 vulns, with 2 crits and three highs! In the past, I would have been saddened by such a thing. But now, I actually believe this is a harbinger of a safer, more secure world. Sure, it’ll take some time for us to get there as we swim through the swamp of vulns and have all kinds of difficulties deploying them and fighting off attackers. I think we’ll be seeing a BIG uptick in the number of vulnerabilities disclosed and patched by most vendors as they leverage AI models to analyze their source code. And then, the vuln numbers will start going down, in say a year or so, and we’ll have a safer, less vulnerable base of software in our environments.
- Slide 2 of 3
As these flaws can be exploited over HTTP without authentication, you want to upgrade to FortiSandbox 5.2 now. Don’t get tunnel vision on the sandbox and miss the other alerts for FortiDDoS, FortiClientEMS and FortiAnalyzer Cloud. Fortinet devices and services are already on attackers’ radar, so don’t be the victim of their curiosity. Also, make sure that you’re restricting access to these services; don’t make the attack any easier than it has to be.
- Slide 3 of 3
Fortinet’s software development practices continue to be called into question given the number of vulnerabilities discovered over the last 18-24 months. Bottom line: if you use Fortinet products, pay close attention to issued product advisories and patch as soon as software updates are made available.
Slide 1 of 0- Slide 1 of 3
Cisco Fixes Critical Flaws in Webex Services, ISE, and ISE Passive Identity Connector
This week, Cisco, published 10 security advisories to address security issues in Cisco Webex Services, Cisco Webex Contact Center, Cisco Secure Web Appliance, Cisco Identity Services Engine (ISE), Cisco Unity Connection, and Cisco ThousandEyes Enterprise Agent. In all, the bulletins address 15 security issues, including four critical vulnerabilities: a vulnerability in the integration of single sign-on (SSO) with Control Hub in Cisco Webex Services (CVE-2026-20184) that could be exploited to allow an unauthenticated, remote attacker to impersonate any user within the service; a remote code execution vulnerability in Cisco ISE and ISE Passive Identity Connector (ISE-PIC, CVE-2026-20147); and a path traversal vulnerability (CVE-2026-20180) and a command injection vulnerability (CVE-2026-20186) in Cisco ISE.
Editor's Notes
- Slide 1 of 3
The fix for the Webex SSO flaw is pretty easy, essentially updating the SAML certificate used. Get that done and move onto ISE: there are no workarounds, so you need to update.
- Slide 2 of 3
Patching is difficult; it requires knowledge, skill, ability, experience, dedication, and courage. It should be recognized and compensated accordingly.
- Slide 3 of 3
I know it sucks, but get used to patching your critical infrastructure. It’s going to become irrelevant whether it’s Cisco, Juniper/Aruba, or whoever. The number of bugs is about to go up dramatically, and knowledge of network equipment is becoming increasingly easy to come by.
Slide 1 of 0- Slide 1 of 3
The Rest of the Week's News
Mythos Preview Tested in CTFs, Cyber Ranges
The UK's AI Security Institute (AISI) has conducted independent testing on the autonomous cyberattack capabilities of Anthropic's Mythos Preview model, comparing its performance to contemporary models. Testing skills in isolation within capture-the-flag (CTF) challenges, Mythos Preview's average success rate was the highest among the group on Expert level (73%) and Apprentice level tasks, and closely followed other models on Technical non-expert and Practitioner level tasks. At all four levels of CTF, Mythos did not exceed its nearest competitors' success rates by more than ten percentage points, and no competitor exceeded Mythos by more than five percentage points. However, Mythos is the first model to fully solve "The Last Ones," (TLO) AISI's 32-step cyber range simulating an attack on a corporate network, "spanning initial reconnaissance through to full network takeover, which [AISI] estimate[s] to require humans 20 hours to complete." Across ten attempts, Mythos averaged 22 completed steps and in three cases solved all 32; the next best performance was from Claude Opus 4.6 at an average of 16 completed steps, with its best attempt reaching 28. In an OT-focused "Cooling Tower" range, Mythos Preview "got stuck on IT sections" and did not complete the challenge. The ranges lack active defenders and defensive tooling, and they do not penalize models for actions that would trigger security alerts, but AISI posits that Mythos Preview "is at least capable of autonomously attacking small, weakly defended and vulnerable enterprise systems where access to a network has been gained." AISI's blog post states that this test "highlights the importance of cybersecurity basics, such as regular application of security updates, robust access controls, security configuration, and comprehensive logging. [...] Future frontier models will be more capable still, so investment now in cyber defence is vital."
Editor's Notes
- Slide 1 of 6
While the capabilities of Mythos and other newly released (and soon-to-be released) models are impressive, there is a lot of talk among researchers of their very high token costs. Such costs may inhibit vulnerability discovery of the have-nots, while simultaneously arming rich organizations with either zero-days for attack, or vuln discovery and patches for defense. The good news here is that current, pre-Mythos models are actually pretty decent at vulnerability discovery. Our team and numerous others have tested the (sort of) affordable current models at open source software projects with some major vulnerability discoveries and patches being issued. Such activity has and will continue to help get rid of a certain level of flaws before the new generation of models become commonplace (and maybe drop in cost).
- Slide 2 of 6
SKEWING OLD FOR FIRST FEW SENTENCES: A bit over 30 years ago, Dan Farmer released the Security Administrator Tool for Analyzing Networks (SATAN) that was essentially one of the first automated vulnerability scanner tools — causing a very similar uproar about bad guys using it. Of course, they did, leading to faster attacks — but so did good guys, leading to faster remediation. In this case, bad guys can use AI to find vulnerabilities and rapidly create attacks, and software developers should be able to use the same technology to more rapidly (as in before releasing bad code) create hardened versions of code. This never-ending race is why AISI’s post ends essentially with "essential security hygiene is required."
- Slide 3 of 6
As good as Mythos is, we still need skilled red teamers to prompt and guide it. The trick is gaining those skills without deferring to AI. Our defenders need tools that help them rapidly assess and respond to deficiencies, as well as cyber hygiene. The old model of hard on the outside and soft on the inside must become a thing of the past. Check out your boundary, make sure you really understand your architecture, then verify it’s all covered so your defenders can, well, defend it. Lastly, it’s not going to matter how well you patched if reusable credentials are compromised.
- Slide 4 of 6
Finding vulnerabilities using LLMs is really interesting, but the really exciting part is figuring out what risk this actually poses. From AISI’s testing, this potential flood of *new* vulnerabilities represents more examples of classes of problems we already know about, where existing security guidance from organizations like the Center for Internet Security still applies. As AISI summarized, they cannot say whether Mythos would succeed against a well-defended system. Essential cyber security safeguards (CIS Controls IG1) can still be an effective defense.
- Slide 5 of 6
I think Mythos is one of the first models that we will see using the latest cards and processes. Most AI models trained on the latest learning hardware will likely follow and best these. What we need to focus on is not that this is happening, but how to handle this. In my opinion, the only way to handle this is to get on the AI train for offensive operations if you’re on that side of the house. If you are on the defensive side, time to get on that train as well. Let the agents help you find what you're looking for. Give it hands.
- Slide 6 of 6
Think "least privilege" and "defense in depth."
Slide 1 of 0- Slide 1 of 6
NIST Triages CVE Enrichment, ENISA Earning TL-Root CNA
In response to a 263% surge in CVE submissions between 2020 and 2025, the US National Institute for Standards and Technology (NIST) has announced their shift toward a "new risk-based model" for prioritizing CVE analysis in the National Vulnerability Database (NVD). Rather than enriching every new CVE — which includes adding associated information such as severity score, common weakness enumeration (CWE), and common platform enumeration (CPE) — after April 15, 2026, NIST will prioritize three categories for enrichment: 1. CVEs added to the Known Exploited Vulnerabilities catalog maintained by the Cybersecurity and Infrastructure Security Agency (CISA KEV), 2. CVEs affecting software used in the federal government, and 3. CVEs for software designated as critical under Executive Order 14028. Users can email nvd@nist[.]gov to request enrichment of unscheduled CVEs. Backlogged CVEs with an NVD publish date prior to March 1, 2026 will not be scheduled for enrichment, with the exception of those in the CISA KEV. Meanwhile, the European Union Agency for Cybersecurity (ENISA) is in the process of becoming the third top-level root CVE Numbering Authority (TL-Root CNA), expecting to join CISA and MITRE in this role within the next year.
Editor's Notes
- Slide 1 of 2
One approach to clearing a backlog is to descope those efforts, and to prevent recurrence, NIST is also reducing their scope of enrichment to something they feel they can maintain. The NIST team remains flat at 21, regardless of volume, so they had to do something. Descoping also means that non-enriched CVEs will only have the CVSS score they were submitted with, if any. Last year CISA stepped in to aid with enrichment; one wonders if it’s possible to have other entities take on the enrichment as well.
- Slide 2 of 2
While the shift to a risk-based model defined by threats to the federal government may make sense on paper, the reality is that the US is stepping back in its role in the vulnerability awareness ecosystem. The timing of the release and use of Claude Mythos and Chat GPT-5.4-Cyber by software vendors may help. It has the potential to reduce vulnerabilities in the development cycle before they become a security problem. We’ll see.
Slide 1 of 0- Slide 1 of 2
CISA Warns Users Who Haven’t Updated Excel in 17 Years
A critical flaw has been added to the US Cybersecurity and Infrastructure Security Agency's Known Exploited Vulnerabilities catalog (CISA KEV), affecting Microsoft Office Excel 2000 SP3, 2002 SP3, 2003 SP3, and 2007 SP1; Excel Viewer 2003 Gold and SP3; Excel Viewer; Compatibility Pack for Word, Excel, and PowerPoint 2007 File Formats SP1; and Excel in Microsoft Office 2004 and 2008 for Mac. 17-year-old CVE-2009-0238 carries a CVSS score of 8.8, and allows a remote attacker to execute arbitrary code when a targeted user opens a crafted Excel document that triggers an access attempt on an invalid object. CISA has not offered details about current exploitation, but the original CVE entry notes that the flaw was already exploited to deliver a trojan dropper at the time of disclosure in 2009. Federal Civilian Executive Branch (FCEB) agencies have until April 28 to remediate this flaw.
Editor's Notes
- Slide 1 of 3
The time has come to address use of old, unsupported Office versions head on. Consider Office 2024 where you want a standalone license. If you're dealing with the old OS, old office use case, verify whether the combination is still required and whether the office component can be moved to a newer system, otherwise you’re going to have to take steps to protect and isolate, to include disallowing internet access by those systems.
- Slide 2 of 3
This may be a first: a 17-year old vulnerability, for which a patch exists, making its way on the KEV list. I would hope that enterprises have upgraded their software long ago; it is after all a critical security control (CSS 2.2, Ensure Authorized Software is Currently Supported) and the basis for demonstrating reasonable cybersecurity.
- Slide 3 of 3
If you haven’t updated Excel in 17 years, I highly doubt that this one CVE is going to be the most important one. I wonder who is making this a KEV, because this seems like kind of sort of a waste of an entry. Is it under active exploitation? Is there something I’m missing here? They are probably also running Windows 8.
Slide 1 of 0- Slide 1 of 3
Microsoft Announces Extended Security Updates for Exchange Server 2016 and 2019
Microsoft is offering users still running Exchange 2016 and 2019 a second period of Extended Security Updates (ESU), ostensibly to allow them additional time to migrate to Exchange Subscription Edition (SE). The first ESU period began in October 2025 and ends this month. The new ESU period will begin in May 2026 and will run through October 2026. Microsoft Enterprise Agreement (EA) customers wishing to enroll in the second ESU period must purchase a new contract to receive Critical and Important security updates. Microsoft is offering a similar ESU program for Skype for Business 2015 and 2019.
Editor's Notes
- Slide 1 of 3
Are you sure you still need to run on-premises Exchange or Skype? While this support is intended to provide more time and incentive to move to the SE version, and this ESU isn’t all roses, it has caveats. I've been in the "we need more time to migrate" discussions; Microsoft is making the delay more uncomfortable. It may be better overall to get out of running these services locally. Have the conversation.
- Slide 2 of 3
MSFT is acknowledging that there is still a large install base for these end-of-life products. One would have thought that affected organizations would have planned for the upgrade over the past ten years.
- Slide 3 of 3
Turns out not everyone wants to go to Microsoft 365, or as they now call it, Microsoft 365 Copilot App. I have to imagine that at some point, Microsoft will either force everyone off or an alternative will appear.
Slide 1 of 0- Slide 1 of 3
Tennessee Hospital Informs Patients of Data Compromise
Cookeville Regional Medical Center (CRMC) in Tennessee has notified 337,917 patients that a July 2025 ransomware attack compromised their personal and protected health information (PHI). CRMC became aware of the attack on July 14, 2025, and a subsequent forensic investigation determined that the intruder had access to CRMC systems between July 11 and July 14, 2025. CRMC notified the US Department of Health and Human Services Office for Civil Rights (HHS OCR) of the incident in August 2025, using a figure of 500 as a placeholder for number of individuals affected. The incident review was completed in mid-March 2026, at which time CRMC obtained a list of affected individuals. The compromised data include addresses, dates of birth, Social Security numbers, driver’s license and financial account numbers, medical treatment information, medical record numbers, and health insurance policy information.
Editor's Notes
That is a big change from the original estimate of 500 impacted individuals. Cookeville is contacting affected individuals and offering credit monitoring services and advice. We’ve got to get faster at the analysis; while this only took 8 months, it doesn’t take nearly that long to exploit pilfered information. The Rhysida ransomware gang took credit for the attack and has leaked 70% of the data, indicating there was a buyer for 30%. Don’t wait to figure out which bucket your data is in; protect yourself now.
Two Men Sentenced for Roles in Fraudulent Remote IT Worker Scheme
A US District Court Judge in Massachusetts has sentenced two people to prison for their roles in facilitating North Korean remote information technology (IT) workers posing as US residents to obtain work at more than 100 US companies. Kejia Wang pleaded guilty to conspiracy to commit wire fraud, conspiracy to commit money laundering, and conspiracy to commit identity theft in September 2025; he was sentenced to nine years in prison. Zhenxing Wang pleaded guilty to conspiracy to commit wire fraud and conspiracy to commit money laundering; he was sentenced to nearly eight years in prison. Kejia Wang traveled to China to confer with contacts involved in the scheme and supervised facilitators of the operation in the US. Zhenxing Wang set up laptops for remote access and enabled technology, including keyboard-video-mouse (KVM) switches that made it appear as though the remote computers were operating within the US. The men were also ordered to forfeit the $600,000 they received for facilitating the operation. Eight additional defendants who were indicted in June 2025 remain at large.
Editor's Notes
Even with these convictions, we still need to be diligent and verify remote workers. I remain a fan of meeting them in person regularly, conducting background checks, mandating phishing-resistant authentication, and providing secure managed environments for remote workers. You’re trying to thwart the use of stolen identities, installation of malware, credential compromise, and unauthorized data exfiltration.