CyberWire Daily - Trust lags behind technology.

Episode Date: April 21, 2026

Anthropic’s Mythos proves irresistible despite claimed supply chain risks.Iran claims U.S. backdoors hit its networks. New Coast Guard rules target maritime OT security. A fresh NGate Android malware variant emerges. Thousands of ActiveMQ servers face active exploitation risk. CISA adds eight flaws to its KEV list. Progress patches MOVEit and LoadMaster bugs. Attackers impersonate IT staff over Microsoft Teams. A ransomware negotiator admits working with BlackCat. Google Gemini asks, “May we see your photos please?” Remember to leave us a 5-star rating and review in your favorite podcast app. Miss an episode? Sign-up for our daily intelligence roundup, Daily Briefing, and you’ll never miss a beat. And be sure to follow CyberWire Daily on LinkedIn. CyberWire Guest On today’s Industry Voices Elad Koren, Vice President, Product Management, Cortex Cloud at Palo Alto Networks, discusses building AI natively into platforms, managing complexity and trust, and taking a measured, experimental approach during the industry’s “messy middle” phase. If you enjoyed this conversation, tune into the full interview here. Selected Reading The US NSA is using Anthropic's Claude Mythos despite supply chain risk (Security Affairs) Anthropic secretly installs spyware when you install Claude Desktop (That Privacy Guy) Iran claims US used backdoors in networking equipment (The Register) Maritime Cybersecurity Rules Make Waves (GovInfoSecurity) New NGate variant hides in a trojanized NFC payment app (We Live Security) Actively exploited Apache ActiveMQ flaw impacts 6,400 servers (Bleeping Computer) CISA flags another Cisco Catalyst SD-WAN Manager bug as exploited (CVE-2026-20133) (Help Net Security) Progress Patches Multiple Vulnerabilities in MOVEit WAF, LoadMaster (SecurityWeek) Microsoft: Teams increasingly abused in helpdesk impersonation attacks (Bleeping Computer) Florida Man Working as a Ransomware Negotiator Pleads Guilty to Conspiracy to Deploy Ransomware and Extort U.S. Victims (United States Department of Justice) Google Starts Scanning All Your Photos As New Update Goes Live (Forbes) Share your feedback. What do you think about CyberWire Daily? Please take a few minutes to share your thoughts with us by completing our brief listener survey. Thank you for helping us continue to improve our show. Want to hear your company in the show? N2K CyberWire helps you reach the industry’s most influential leaders and operators, while building visibility, authority, and connectivity across the cybersecurity community. Learn more at sponsor.thecyberwire.com. The CyberWire is a production of N2K Networks, your source for strategic workforce intelligence. © N2K Networks, Inc. Learn more about your ad choices. Visit megaphone.fm/adchoices

You're listening to the Cyberwire Network, powered by N2K. No, it's not your imagination. Risk and regulation really are ramping up, and these days customers expect proof of security before they'll even do business. That's where Vanta comes in. Vanta automates your compliance process and brings compliance, risk, and customer trust together on one AI-powered platform. So whether you're getting ready for a SOC2 or managing an end-toe, enterprise governance risk and compliance program, Vanta helps keep you secure and keeps your deals

moving. Companies like Ramp and Writers spend 82% less time on audits with Vanta. That means less time chasing paperwork and more time focused on growth. For me, it comes down to this. Over 10,000 companies from startups to large enterprises trust Vanta to help prove their security. Get started at vanta.com slash cyber. Anthropics mythos proves irresistible despite claimed

supply chain risks. Iran claims U.S. back doors hit its networks. New Coast Guard rules target maritime OT security.

A fresh NGate Android malware variant emerges. Thousands of active MQ servers face active exploitation risk. Sisa adds

eight flaws to the known exploited vulnerabilities list. Progress, patches, Move It, and Loadmaster

bugs. attackers impersonate IT staff over Microsoft teams. A ransomware negotiator admits working with Black Cat. Our guest is Elad Koren, Vice President of Product Managent for Cortex Cloud at Palo Alto Networks. He's discussing building AI natively into platforms. And Google Gemini asks, may we see your photos, please? It's Tuesday, April 21st, 2026.

I'm Dave Bittner, and this is your Cyberwire, Tell Briefing. Thanks for joining us here today. It's great as always to have you with us. Axios reports that the U.S. National Security Agency is using Anthropics Mythos preview model, even as the Department of

Defense labels Anthropic a supply chain risk, highlighting the growing tension between AI capability and security governance. Mythos is especially strong at finding software vulnerabilities, which makes it valuable for cyber defense, but

potentially useful for offensive activity if misused. This dual-use nature reflects a broader policy challenge. Agencies may depend on advanced AI tools despite concerns about vendor trust, oversight, and strategic reliance. Anthropics-related Project Glasswing initiative aims to apply mythos defensively with industry partners to secure critical infrastructure. The episode underscores how operational urgency is pushing governments to adopt frontier AI faster than procurement rules and risk frameworks can adapt. Elsewhere, security researcher Alexander Hemp reports that installing Anthropics Clodd desktop on Mac OS

automatically deployed an undocumented native messaging bridge across multiple chromium-based browsers without user notice or consent. The manifest allows approved browser extensions to launch a local helper binary outside the browser sandbox with user-level privileges, potentially enabling access to authenticated sessions, page content, and automation features if activated. Hanf says the files were installed even for browsers not present on the system and were repeatedly reinstalled after deletion. Anthropics documentation describes similar capabilities for its browser integration but does not document this specific bridge behavior. The author argues the design expands a tax surface and weakens browser isolation,

raising concerns about transparency, consent, and supply chain trust in desktop AI integrations. Iranian media claims U.S. intelligence used hidden back doors or botnets to, disable networking equipment from vendors, including Cisco, Juniper, Fortinette, and Microtic during recent wartime disruptions, even while Iran was largely disconnected from the global Internet. The allegations suggest firmware-level sabotage triggered remotely, though the outages are difficult to verify due to Iran's ongoing nationwide connectivity restrictions. Some reports instead proposed compromise devices formed part of a pre-positioned botnet, Chinese state media has amplified the claims, citing them as evidence of long-standing assertions

that the United States embeds surveillance capabilities in networking infrastructure. Meanwhile, monitoring group net blocks reports Iran's internet blockade has lasted more than 50 days with selective access granted to favored users, underscoring the limited visibility into events on the ground. A new U.S. Coast Guard rule requiring cybersecurity controls for operational technology in ports and for large U.S. flagged vessels is expected to significantly expand the maritime cybersecurity market amid rising geopolitical risk. Operators must appoint cybersecurity officers, conduct risk assessments, develop vessel and facility cybersecurity plans by July 27, and comply with incidents. reporting and training requirements already in effect. Industry experts say the rule will help security teams justify funding, though guidance on implementation details remains limited.

The Coast Guard estimates compliance will cost about $134 million annually, a large share of a maritime cybersecurity services market valued at $186 million globally in 2024. While larger firms may build internal capabilities, abilities, smaller operators may rely on outsourcing. Enforcement capacity remains uncertain as the Coast Guard prepares to integrate cyber checks into existing inspections. Researchers at ESET identified a new NGATE Android malware variant that trogenizes the legitimate handy pay app to relay payment card NFC data and steal pins for fraudulent ATM withdrawals and transactions. The campaign, active

since November of last year, targets users in Brazil through fake lottery and counterfeit Google Play websites distributing the modified app. Researchers say the injected malicious code shows signs of possible generative AI involvement, though this remains unconfirmed. Like earlier NGate attacks using tools such as NFCGate, attackers modified handi-paid directly to reduce cost and suspicion. The malware forwards card data to attacker devices and exfiltrates pins to a command and control server, reflecting a broader rise in NFC-enabled financial fraud. More than 6,400 Internet-exposed Apache Active MQ servers are vulnerable to active exploitation of a high-severity code injection flaw, according to Shadow Server.

The issue, caused by improper input validation, allows authenticated attackers to execute arbitrary code and was reportedly identified with assistance from the clawed AI tool after remaining undiscovered for 13 years. Adobe patched the flaw in March, but many systems remain unpatched. SISA has ordered federal agencies to remediate the risk by April 30th. Speaking of SISA, they've added eight vulnerabilities to the known exploited vulnerabilities catalog, including three affecting Cisco Catalyst SDWAN manager, which Cisco confirmed as exploited.

The update also includes flaws in paper cut, NGMF, JetBrains Team City, Kentico Experience, Quest Case Appliances, and Sinakor-Zimbra Collaboration Suite, some link to ransomware and espionage activity. Federal agencies must remediate all eight by April 20th. Progress software released patches for multiple MoveIt web application firewall and loadmaster vulnerabilities that could allow authenticated attackers to execute arbitrary commands through improperly sanitized API inputs and file uploads. The flaws affect administrative functions in Progress ADC products. Another issue enables specially crafted requests to bypass firewall protections. Progress says there are no reports of active exploits.

but urges customers to update affected systems promptly. Microsoft warns threat actors are abusing external Microsoft Teams chats to impersonate IT or help desk staff and trick employees into granting remote access to enterprise systems. In observed campaigns, attackers initiate support sessions using tools like Quick Assist, then perform reconnaissance with command prompt and PowerShell, established assistance through DLL side-loading and move laterally using Windows remote management. They deploy additional remote tools and use utilities such as R-Clone to selectively exfiltrate sensitive data to cloud storage. Because the activity relies heavily on legitimate

software and native administrative protocols, detection is difficult. Microsoft advises organizations to treat external teams' messages as untrusted and restrict remote assistance and WIN-R-M usage to reduce risk. A Florida man and former ransomware negotiator, Angelo Martino, pleaded guilty to conspiring with Black Cat ransomware operators to target U.S. companies in 2023. Prosecutors say Martino abused his role at a cyber incident response firm to share victims' confidential negotiation strategies

and insurance limits with attackers, helping increase ransom payments. He also joined accomplices Ryan Goldberg and Kevin Martin in deploying Black Cat ransomware against multiple victims, including an attack that yielded about $1.2 million in Bitcoin. Authorities have seized more than $10 million in assets linked to the scheme. Martino faces up to 20 years in prison with sentencing scheduled for July. Officials say the case highlights insider risk

within the ransomware response ecosystem. Coming up after the break, my conversation with Elad Koren from Palo Alto Networks. We're discussing building AI natively into platforms. And Google Gemini asks, may we see your photos, please? Quick question. Have you watched Project Hail Mary yet?

Humanity is facing an existential threat and racing to solve it with the clock ticking. For security teams, that probably hits close to home with AI use rapidly spreading. Everyone's using AI, marketing, sales, engineering. Chris the intern without security even knowing about it. That's where Nudge Security comes in.

Nudge finds Shadow AI apps, integrations, and agents on day one and helps you enforce policy without blocking productivity. Try it free at Nudgesecurity.com slash cyberwire. Maybe that's an urgent, message from your CEO, or maybe it's a deep fake trying to target your business. Dopple is the AI-native social engineering defense platform fighting back against impersonation

and manipulation. As attackers use AI to make their tactics more sophisticated, Dopple uses it to fight back, from automatically dismantling cross-channel attacks to building team resilience and more. Doppel, outpacing what's next in social engineering. Learn more at Dopple.com. That's do P-P-P-E-L.com. Elad Koren is vice president for product management with Cortex Cloud at Palo Alto Networks.

I caught up with him at the 26 R-SAC Conference for this sponsored industry voices segment discussing building AI natively into platforms. I just heard a story earlier today of someone that had a very reliable system built

on an agent that can train for them, but it actually invented some tickers that never existed, right? Well, so how can you prevent something like that? You create the right barrails and the right agents that will analyze this, and you need a system, and this is a real building. Well, here we are at RASAC 2026, and it's my pleasure to welcome to the program, Elad Koren, who is vice president for product management,

And Cortex Cloud at Palo Alto Networks, a lot. Thanks so much for joining us. Thank you for having. Before we dig in here, how's the week going for you with the show? I'd ask, are you keeping busy? But that's sort of a rhetorical question. Yeah, so many people here, so many meetings, interesting meetings with a lot of interesting people. I think all in all, this is the biggest week for the cyber industry.

And definitely something we've. fuel as well. Yeah. I want to dig in with you on AI, which of course is the hot topic here this year. You may have notice. But particularly agentic AI, which is the hottest of the hot parts of the hot topic this year. First of all, just from a really high level, how are you and your colleagues approaching this transition to the dominance of a in the conversation.

Yeah, I think one is evident. It's everywhere. It's in a conversation with customers, organizations, a lot of other vendors that are throwing this into the mix. I think

what we're seeing now is a race. A race to who can add AI to their solution, the fastest way. We are the way we are thinking about it And I think, you know, there are many approaches. Naturally, we think ours is the right one to address this.

We all agree at the baseline for everyone that you cannot really fight machines with only humans. You need machines to come to the assistance of humans when fighting with machines, which are the adversaries, the sprawl of AI, the hand of those that are trying to infiltrate, breach information, et cetera. I think what we're seeing is that the best approach, and this is what resonates with many of

the organizations we are having conversations with, is that when we built specifically in the vortex platform, when we built the platform itself, we thought of AI as an inherent part of the platform. It's not something

bold to go. Okay. And what it helps with is the access to the data, the lake, everything exists there. You don't need to reach out to other systems. Everything is right there and available. But also, it helps a lot with

multiple agents that are trying to do different things, right? Because we're, as humans, we are thinking about the, you know, we are optimistic features, right? We have to. So, everyone are saying, great, when we have a Jetty AI, all of our problems

will be solved. Oh, guess what? We were adding complexity. We're not just removing the complexity. So our agents from different vendors or different tasks or jobs to be done are going to interact with each other. Are they really going to imagine a kitchen with so many cooks? What happens? Somebody gets burnt. So we are, the way we're thinking about it is, you know, AI needs to be an integral part of the solution so that all the agents can be an integral part of the solution so they're orchestrated well and working well with each other. This is our philosophy going at it, and we see that it resonates.

How is that playing out? I mean, it strikes me as something that could be easier said than done to place appropriate guardrails on there. I can imagine all the agents vying to be the one, the alpha. Well, yes or no. When you design the agents in a way that, one, bleats back and relies on the purpose. people or the employees, the humans that created them. And they have the right guardrails in place to make sure that they cannot do things that are more than what the human created them can do.

So this is basic. But also, they have a task. They have a potential journey that they need to take. And if you have two agents, one needs to take care of one thing and the other needs to take of something else. they operate as two humans. And if the system is designed to look at potential collusions or potential cases where they may contradict,

you think about what should happen when such two, three, four, maybe ten regions do on the system and how they interact with data, and you plan for this. If you plan for this and you design the system in such a way, then it definitely solves the problem in advance. Does it remove all problems? Probably not, right? But I think this is, this is more.

I like to call this time when we are in the messy middle. We are in the messy middle. We are discovering that as we go along. If somebody says that they know how it's going to look in three, four, five years, well, either they have a crystal ball or their life. So I think we're discovering this. Yes, we are solving for those. We are as an industry solving for those specific cases.

But I think generally speaking, there's a good understanding. of how the fundamentals work. So it sounds to me like you're optimistic about the direction that we're headed and we're going to be able to strike that necessary balance?

Yes, yes. I think, and it has a few aspects to it. One, we still have a way to go as it pertains to trust, right? Let just like people would go into a waymo,

10 years ago? Something changed. They trusted. It's the same with having autonomous systems. You cannot just jump at the autonomous system. You have to gain this trust. You have to start with a smaller, simpler things. You then have to grow up and be more at more complex stuff.

Well, one or two or three agents collaborate and bring to a resolution or mediation of something, It can be cloud. It can be sock. Whatever, right? And only then you have that trust in the system. So this is one aspect. But the other thing is the capabilities.

Many of the capabilities today are not quite there. We see hallucinations. We see, you know, I just heard a story earlier today of someone that had a very reliable system built on an agent that can trade for them, but it actually invented some tickers that never existed, right? Well, so how do you prevent some? something like that, you create the right guardrails and the right agents that will analyze this. And you need a system.

And this is what we're building. So it's trust, it's the right infrastructure, it's the right capabilities, it's the right guardrails. And when you start doing all of those, you cannot design them from the get-go. Perfect. It's never the case. We have to discover the problems, solve the problems, discover more problems, solve those problems and move on. This is how we build complex systems, right? Yeah, what's your advice to that person who says, I am nervous about this?

I seize the potential problems. But at the same time, I'm afraid that if I don't jump on board, my competitors are going to outpace me. It's just a fact and certain. However, I think the responsible thing to do is to decide on the right risk appetite. I can definitely see a future where someone, a vendor, that is staying still not really hasting to put in AI their solutions in waiting for like six more months. Maybe that will prove to be the right strategy. Let's take a different lens to look at that, that specific question. If you try to do vibe coding a year ago with just Jet Jik-T or any other vendor out there, would it be successful?

Probably not. Probably you'll get a lot of not that great code and probably very messy. But then you waited and now you have so many vibe code. solutions. I don't know if one of our viewers or listeners have tried to use one of the platforms to create an app

with vibe coding, about a year ago, it's very different that it is today. So if someone waits, so vendor waits, you know, they're concerned, great. Do your diligence. Learn. Experiment, explore. I always tell all of my teams, if you

want to be at the the edge of technology, you have to experiment. We're not pushing that into production. But try it. If you don't try, you don't understand the new boundaries. Boundaries have changed. So this is my advice. Try everything

that you can, experiment. Do not add it to your product just yet until you know what purpose it serves. Because if you're just adding AI for the sake of adding AI, it means nothing. The customers will call it out. But if you're adding something that has true benefit,

either cutting some of the costs, bringing solutions and bringing benefit better to the customers, then you should know what it does, right? And you should do the right, both education as well as learning in the process so that you can plan it properly. And I think that is the best thing that organizations, not to jump into this just like for the sake of a half. Yeah.

Well, Al-a-a-a-law, thank you so much for joining us. Elad Koran is vice president for product management, for Cortex Cloud at Palo Alto Networks. Thanks so much. Thank you. That's Elad Corrin from Cortex Cloud at Palo Alto Networks. Local news is in decline across Canada, and this is bad news for all of us. With less local news, noise, rumors, and misinformation fill the void, and it gets harder to separate truth from fiction.

That's why CBC News is putting more journalists in more places across. Canada, reporting on the ground from where you live, telling the stories that matter to all of us, because local news is big news. Choose news, not noise. CBC News. This episode is brought to you by Tell Us Online Security. Oh, tax season is the worst. You mean hack season?

Sorry, what? Yeah, cybercriminals love tax forms. But I've got Tellus Online Security. It helps protect against identity theft and financial fraud, so I can start. stress less during tax season, or any season. Plans start at just $12 a month. Learn more at talus.com slash online security. No one can prevent all cybercrime or identity theft.

Conditions apply. Hey, ladies and gentlemen, guess what? You came on a great night. Tonight is, may we see your photos, please? There you go. And finally, Google's latest Google Photos update invites users to let Gemini browse their memories in the name of convenience and possibly creativity.

By opting in to its new personal intelligence feature, users allow the AI to scan photos of friends, family, and life events, so it can generate more personalized images without needing detailed prompts. In theory, this saves time. In practice, it means your camera roll, that quiet archive of vacations, receipts, pets, and accidental screenshots, becomes part of Gemini's working knowledge. Google says the system does not directly train models on private photo libraries,

though limited prompt and response data may still be used to improve performance. The feature is optional, adjustable, and launching first in the United States. Still, it quietly reframes the trade-off. Fewer instructions for the AI, more access to your life. As always, with convenience upgrades, the fine print arrives slightly after the excitement. All right, thank you very much for playing. May we see your photos, please. And that's the Cyberwire.

For links to all of today's stories, check out our daily briefing at thecyberwire.com. We'd love to know what you think of this podcast. Your feedback ensures we deliver the insights that keep you a step ahead in the rapidly changing world of cybersecurity. If you like our show, please share a rating and review. in your favorite podcast app. Please also fill out the survey in the show notes or send an email to Cyberwire at N2K.com.

N2K's lead producer is Liz Stokes. We're mixed by Trey Hester with original music and sound design by Elliot Peltzman. Our contributing host is Maria Vermazes. Our executive producer is Jennifer Ivan. Peter Kilty is our publisher, and I'm Dave Bittner.

Thanks for listening. We'll see you back here tomorrow.