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
Anthropic Previews Mythos Model to Limited Partners
On Tuesday, April 7, 2026, Anthropic announced a preview of a new general-purpose LLM called Claude Mythos, alongside the launch of an industry group called Project Glasswing, focused on implementing the new model defensively ahead of public release. According to Anthropic, during internal testing Mythos was uniquely successful at scanning for and exploit zero-day vulnerabilities, some of which are decades old, across all major operating systems and browsers, including exploit chains of multiple flaws. While previous models Sonnet and Opus 4.6 have a <1% success rate at autonomously developing exploits, Mythos generated successful exploits in 72.4% of trials. Anthropic states that Mythos has identified thousands of vulnerabilities that are likely high and critical-severity, and the company has taken on contractors to manually validate bug reports as part of ongoing responsible disclosure. Anthropic does not plan to make Mythos Preview generally available; use will at first be limited to Project Glasswing, comprising 12 named partners and ~40 other organizations "that build or maintain critical software infrastructure," including Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. Anthropic has chosen to be reticent about findings in the name of safety, but does detail a few notable discoveries, all of which have been disclosed and patched: a 27-year-old stack-based buffer overflow vulnerability in FreeBSD (CVE-2026-4747); a 16-year-old vulnerability in FFmpeg; and a chain of flaws allowing privilege escalation in the Linux kernel.
Editor's Notes
- Slide 1 of 4
Look at this list of initial partners that are getting to use this tool. I suspect the number of bugs that will be patched over the next few months will be substantial. Now think about the number of vendors that are NOT included in this list. They will need a lot of help, and quickly. If you are running software from these companies and can start looking for bugs, I would strongly suggest it. Aside from cost, this year will be a big one for bugs. The pricing of the model is going to be a major deterrent, as it appears that both Mythos and GPT-5.4 are substantially more expensive for now.
- Slide 2 of 4
During testing, Mythos escaped from its isolated sandbox and was able to email the tester. The point is that if you’re testing this type of AI, you may want physical isolation (think air gap) until you understand its capabilities.
- Slide 3 of 4
We knew this day was coming, perhaps a bit sooner than expected. Claude Mythos excels at the ‘finding’ of vulnerabilities, so time to double down on the ‘fixing’ and ‘installing’ patch components. Collectively we have to find a way to automate patching; otherwise, the fear mongering is well placed. While Glasswing is a good start, the industry group should be widened to include the next level of software providers, including system integrators. This should occur well before release.
- Slide 4 of 4
The findings of Preview should not surprise anyone; Microsoft has been finding tens of vulnerabilities per month for years, and there are hundreds of entries in the KEV. The speed and efficiency of Mythos is a two-edged sword. It promises to exponentially increase the efficiency of testing and maintenance. On the other hand, it has the potential to lower the cost and increase the rate of attack against our vital infrastructure. Even if the rogues do not gain access to Mythos, they will most certainly gain access to similar capability. We can expect Project Glasswing to give us at least a few weeks to clean up our shoddy code base. It is to be hoped that we will take advantage of it. Finally, while the publicity highlights the potential for Mythos to identify exploits, AI can as easily produce fixes and patches. While we can automatically apply fixes to the source, patching all the instances will be a huge effort. Fortunately, we will be able to repair instances faster than the rogues can exploit them. The end result must be a more robust infrastructure; there is no choice.
Slide 1 of 0- Slide 1 of 4
December 2025 Eurail Breach Exposed Passports
A December 2025 breach of Eurail B.V.'s customer database resulted in the theft of passport information belonging to more than 300,000 individuals. Eurail, which is based in the Netherlands, sells Interrail and Eurail train passes. The incident was initially disclosed in February 2026, at which time Eurail said the intruders accessed sensitive information, including names, passport details, ID numbers, bank account IBANs, health information, contact details, email addresses, and phone numbers. In a March 27, 2026 breach notification letter sent to affected individuals, Eurail acknowledged that the threat actors exfiltrated data. A filing with the Oregon Attorney General's office places the number of individuals affected by the breach at 308,777.
Editor's Notes
- Slide 1 of 4
These kinds of systems being hit are possibly some of the worst types of attacks. Passport theft will increasingly be an issue until visual/physical presentation of fingerprints and other data is used. Then again, if you have not gone through a port of entry and someone else does, who is the fake person? This is going to be a problem for a while until we sort out the systems. As more and more KYC systems become required, I really hope a third-party-secured service is created to help secure this type of data.
- Slide 2 of 4
As you investigate an incident, you may discover things that require notification; don’t forget to stay on point with messaging and highlight improvements made and steps taken to maintain or restore customer and employee confidence and relationships. When you get to the other side, be sure to understand and acknowledge responders visibly and tangibly.
- Slide 3 of 4
There is a difference between possessing all the identity information encoded on a passport, the most that Eurail might have, and the ability to forge a modern passport. On the other hand, the information in the database will be for sale on the dark web to those engaged in application fraud. Whether or not you are in the Eurail database, be sure that you have locked your credit bureau information, and monitor your accounts for fraudulent transactions or changes.
- Slide 4 of 4
It appears that the perps absconded with data that can cause great harm to the 309K victims. Eurail, and every company for that matter, should revisit data retention policies looking for ways to reduce the amount of personally identifiable information stored; it's a lesson we learn over and over again.
Slide 1 of 0- Slide 1 of 4
Los Angeles City Attorney's Office Breach Exposed LAPD Files
The Los Angeles Police Department (LAPD) has published a statement disclosing that "discovery documents from previously adjudicated or settled LAPD civil litigation cases" were compromised during a breach of a digital storage system at the LA City Attorney’s Office. The incident did not involve LAPD systems. An LA City Attorney’s Office spokesperson said they were alerted to "unauthorized access to a third-party tool used by the City Attorney’s Office to transfer discovery to opposing counsel and litigants" on March 20, 2026. According to the LA Times, the breach compromised 7.7 terabytes of LAPD records.
Editor's Notes
- Slide 1 of 3
Third-party security is more important than ever. It’s easy to forget when things just work and users are happy. Beyond continuous monitoring and assessment, schedule required actions where security configurations are reviewed and adjusted. Then the hard one, data retention. While it’s nice to have everything online and available, it’s also a liability. Develop a plan: archive or destroy. You need a protection plan for your archive, more than just "it’s a read-only copy", particularly if it’s online.
- Slide 2 of 3
Not a lot of info on the "how" of this one, but it sounds like weak authentication on a sensitive access tool was the root cause of unauthorized access. This is a good reminder to periodically check for new tools in use and work to require strong authentication on all sensitive tools in use.
- Slide 3 of 3
Not a lot of deets here, but it serves as a good reminder that third-party applications are to be part of an organizations cybersecurity program. CISOs should include them as a risk register item they regularly report out on.
Slide 1 of 0- Slide 1 of 3
The Rest of the Week's News
German Authorities Name REvil/GandCrab Suspects at Large
Germany's Federal Criminal Police Office (Bundeskriminalamt, BKA) has revealed the identities of two wanted Russian men believed to be former leaders of the GandCrab and later REvil ransomware-as-a-service (RaaS) operations. 31-year-old Daniil Maksimovich Shchukin and 43-year-old Anatoly Sergeevitsch Kravchuk are suspects in at least 130 cases of computer sabotage and extortion between 2019 and 2021, allegedly collecting about €2 million in ransom payments and causing over €35 million of damage. Brian Krebs notes that REvil and GandCrab "pioneered the practice of double extortion — charging victims once for a key needed to unlock hacked systems, and a separate payment in exchange for a promise not to publish stolen data." REvil formed after GandCrab dissolved, and targeted high-profile companies and individuals for extortion before being dismantled by FBI disruptions in 2021 and arrests in 2022. The BKA believes the two men to reside in Russia; both have been added to the EU Most Wanted list.
Editor's Notes
- Slide 1 of 2
UNKN, aka Schukin, gave an interview describing his rags to riches story, not overly encumbered by morals and ethics, promoting the economic gains of getting into the ransomware business. Double extortion and ransomware-as-a-service have turned out to be lucrative, if you ignore that little issue of legality. Having faces and names tied to their handles will hopefully accelerate their apprehension.
- Slide 2 of 2
Naming and placing on most wanted lists is effective in limiting an evildoer’s ability to enjoy the fruit of their crimes. It doesn’t stop the crime, but does put one on notice limiting travel options. Good work, BKA!
Slide 1 of 0- Slide 1 of 2
Ivanti EPMM Critical Flaw Added to CISA KEV with Three-Day Deadline
The US Cybersecurity and Infrastructure Security Agency has added a critical flaw in Ivanti Endpoint Manager Mobile (EPMM) to the Known Exploited Vulnerabilities catalog (CISA KEV). CVE-2026-1340, CVSS score 9.8, allows a remote unauthenticated attacker to achieve code execution due to a code injection vulnerability in Ivanti EPMM. This flaw was originally disclosed on January 29, 2026, simultaneously with another flaw with an identical description (CVE-2026-1281), less than 24 hours before attacks on mobile management infrastructure impacted the European Commission and government organizations in the Netherlands and Finland. CVE-2026-1281 was added to the CISA KEV the same day, but CVE-2026-1340 was not added until April 8. Customers should check the indicators of compromise (IoCs) linked in Ivanti's original advisory and follow the instructions and syntax to properly apply the RPM patch script appropriate to their system. Federal Civilian Executive Branch (FCEB) agencies are required to patch this flaw by February 11.
Editor's Notes
- Slide 1 of 2
Here is another short lead time required update. Added April 8th, due April 11, and looking at my watch, that’s tomorrow. Make sure that the Ivanti patch is already deployed, and that you’re already looking for the IoCs; the patch RPM was released in February.
- Slide 2 of 2
It would be interesting to know if CISA believes that agencies have developed the capability to patch within the three-day mandate.
Slide 1 of 0- Slide 1 of 2
Vehicle Fleet Software Developer Takes Systems Offline
Chevin Fleet Solutions, a UK-based developer of software for managing the details and logistics of vehicle fleets, took its FleetWave platform offline following an "incident" on April 2, 2026, and it remains down as of this writing. The Register reports that an email sent to customers states that Chevin took FleetWave's Azure environments offline as a precaution, and that the company has engaged external cybersecurity experts to investigate and implement security controls. This outage has caused disruptions to users of the software in the US and the UK, though the EU and Australia still have working service. Chevin has not provided details on the nature, timing, scope, or impact of the incident, but their website indicates which components of their service are operational, and directs customers to Zendesk for updates.
Editor's Notes
- Slide 1 of 3
Consider the side effects of withholding incident details. Customers need to understand the underlying risk behind a major shutdown as well as what happened to trigger the event. This will also help them appreciate measures taken to prevent recurrence, particularly if they are impactful.
- Slide 2 of 3
Well, "as a precaution" is better than using the PR-speak "in an abundance of caution," but both mean, "because we are not yet sure of the extent of the problem."
- Slide 3 of 3
Bottom line: Chevin Fleet Solutions is in full damage-control mode. It will be interesting what impact it reports to business operations, including finances, from the extended outage.
Slide 1 of 0- Slide 1 of 3
Signature Healthcare Cybersecurity Incident Disrupts Services
A cybersecurity incident at Signature Healthcare in Brockton, Massachusetts on Monday, April 6, prompted the organization to divert ambulances from Brockton Hospital to other facilities. Signature Healthcare identified "suspicious activity within a portion of [its] network ... [and] activated ... incident response protocols." While the switch to downtime operations has not impacted the organization's surgeries and procedures, chemotherapy infusion services scheduled for Tuesday, April 7 were cancelled, and patients have been notified of possible delays at Signature Medical Group and related urgent care facilities. Some Signature Healthcare retail pharmacies were closed on Monday; when they reopened on Tuesday, April 7, they were able to provide consultations but unable to fill prescriptions. According to a Signature Healthcare update on Wednesday, April 8, lab work and tests are being conducted but may be delayed; requests for medical records are temporarily unable to be fulfilled; ambulance traffic remains diverted; and chemotherapy infusion services have resumed.
Editor's Notes
Signature Healthcare is doing a great job keeping their website updated as their incident response progresses. If you’re a patient, keep an eye on that site, as they are including contact information if you have questions. No threat actors have yet claimed this attack, so it will be interesting to see what the root cause was.
EXPMON Analysis Detects Adobe Acrobat Reader Zero-day
A zero-day vulnerability in Adobe Acrobat Reader has been under exploit since late last year, according to researcher Haifei Li, founder of the EXPMON, an analysis system that "uses crafted sandboxing and static analysis techniques to check if a file sample could be a potential exploit." The Adobe Reader exploit leverages maliciously crafted PDF documents. Li writes that according to EXPMON analysis, " the sample acts as an initial exploit with the capability to collect and leak various types of information, potentially followed by remote code execution (RCE) and sandbox escape (SBX) exploits. It abuses zero-day/unpatched vulnerability in Adobe Reader that allows it to execute privileged Acrobat APIs, and it is confirmed to work on the latest version of Adobe Reader." At the time of writing, Adobe has not released a statement regarding the vulnerability.
Editor's Notes
- Slide 1 of 2
This is one of those developing threat stories. The malicious PDF, which includes a highly obfuscated JavaScript payload, starts by profiling systems, and systems which meet the right (as yet unknown) circumstances get a secondary malicious payload. The blog post from EXPMON includes two addresses to block, but not much else in terms of IoCs. Beyond that, there is currently no CVE nor update; we need to double down on using caution opening PDF documents from unknown sources.
- Slide 2 of 2
It’s been a long time since I’ve had to talk about an Adobe Reader vulnerability.
Slide 1 of 0- Slide 1 of 2
Healthcare Software Company ChipSoft Suffers Ransomware Attack
Dutch healthcare software company ChipSoft was the target of a ransomware attack that has resulted in its systems going offline on Tuesday, April 7. ChipSoft provides Electronic Health Record (EHR) systems to many hospitals in the Netherlands; eleven of those hospitals have taken their ChipSoft software offline. Z-CERT, the country's computer emergency response team, has published an advisory about the attack, noting that "ChipSoft indicates that all connections to the Zorgportaal, HiX Mobile, and the Zorgplatform have been disabled as a precaution and are currently unavailable. ChipSoft has started bringing the systems back online in phases, during which users are receiving new login credentials."
Editor's Notes
- Slide 1 of 2
Dutch healthcare providers are under an increased risk of ransomware attacks, and while this attack hasn’t been categorized as such, it’s the most likely scenario. The good news is that services are being restored quickly, with the largest customer impact being setting a new password. One hopes they are taking the opportunity to implement phishing resistant MFA.
- Slide 2 of 2
ChipSoft is in the supply chain. A successful breach of their systems creates risk for all of their customers. Those customers are in healthcare, the favorite target of extortion attacks. Taking EHR systems offline raises cost for their customers and may result in failure to record important test results and treatments. EHRs are more vulnerable to massive leaks than the paper files that they replace. While it might not be quite true to say that patients are dependent upon them, taking them offline will inconvenience thousands of people. (I would certainly miss mine).
Slide 1 of 0- Slide 1 of 2