CyberWire Daily - The robots have gone bananas.

Episode Date: August 20, 2026

Federal agencies warn of an active campaign targeting critical infrastructure. Citrix races to patch critical NetScaler flaws. More than 50,000 exposed Stripe API keys raise fraud concerns. Black Hat and DEF CON attendees are targeted in a new social engineering campaign. Atlassian, Splunk, and Cisco fix hundreds of vulnerabilities. A new Android banking trojan adds an unusual twist. A healthcare breach impacts 3.8 million people. SilkParasite expands cyberespionage in Central Asia. And CISA eyes a major overhaul of federal cyber software procurement. Our guest is Chris Wallis, founder and CEO of Intruder, on how AI agents killed the annual pentest and are reshaping exposure management. AI powered robots find bananas quite appealing.  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 our Industry Voices segment, Intruder’s Founder and CEO Chris Wallis joined Dave at Black Hat to discuss why the annual pentest Is dead and how AI agents are reshaping exposure management. If you enjoyed this conversation, be sure to check out the full interview here. Selected Reading NSA, FBI warns of hackers using AI-generated tools in attacks on critical infrastructure technology (The Record) Citrix urges admins to patch new NetScaler flaws as soon as possible (Bleeping Computer) 50,000 Stripe Secrets Leaked in Public Code (SecurityAffairs) Black Hat/DEF CON attendees targeted in malware scheme with Google Doc lure (SC World) Atlassian, Splunk Patch Dozens of Critical, High-Severity Vulnerabilities (SecurityWeek) Cisco Patches Critical Crosswork, Secure Workload Vulnerabilities (SecurityWeek) New Manic Android malware can exfiltrate data through nearby devices (Bleeping Computer) EHR Vendor Notifying 3.8 Million Patients of Data Theft Hack (GovInfo Security) SilkParasite: Tracking a China-Nexus APT Across Central Asia (Bitdefender) CISA contemplates whether to hire security software buying help (Washington Technology) I Saw the Future of AI in a Robot That Can Learn on the Spot (WIRED) 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.

You're listening to the Cyberwire Network, powered by N2K. Social engineering attacks look trustworthy, a routine request, an internal email, a familiar face on a call. But Dopple sees through the disguise. Their AI-native platform detects and disrupts attacks across every channel, trains employees to recognize deepfakes and deception, and investigates every fish to take down the campaign behind it. They fight relentlessly to protect your... your business, brand, and people. Dopple, outpacing what's next in social engineering.

Learn more at doppel.com. That's do p-p-p-el.com. Federal agencies warn of an active campaign targeting critical infrastructure. Citrix races to patch critical net scalar flaws. More than 50,000 exposed stripe API keys raise fraud concerns. Black Hat and DefCon attendees are targeted in a new social engineering campaign. Atlassian, Splunk, and Cisco fix hundreds of vulnerabilities. A new Android banking Trojan adds an unusual twist.

A health care breach impacts 3.8 million people. Silk Parasite expands cyber espionage in Central Asia. Sisa is a major overhaul of federal cyber software procurement. Our guest is Chris Wallace, founder and CEO of Intruder on how AI agents killed the annual pen test and are reshaping exposure management. And AI-powered robots find bananas quite appealing. It's Thursday, August 20th, 2026.

I'm Dave Bittner, and this is your Cyberwire Intel briefing. Thanks for joining us here today. It is great as always to have you with us. Federal agencies are warning of an active cyber campaign targeting critical infrastructure through Internet-exposed Siemens S-7 series programmable logic controllers. which are widely used in energy, water, agriculture, manufacturing, and defense. According to a joint advisory from the NSA, FBI, and other agencies,

attackers are using artificial intelligence to generate exploit scripts that accelerate the discovery and exploitation of known vulnerabilities. These AI-assisted tools can mimic legitimate monitoring software, lowering the technical barrier for attackers and enabling faster adaptation to defensive measures. Officials say the activity appears focused on reconnaissance and capability development, potentially laying the groundwork for future disruptive attacks rather than immediate data theft. The advisory urges organizations to isolate PLCs from the Internet,

apply security patches, and strengthen monitoring. Experts warn that many exposed PLCs are connected by third-party events, vendors without operators' knowledge, increasing the risk of operational disruption, equipment damage, and loss of control over critical industrial processes if attackers succeed. Citrix is urging customers to immediately patch two newly disclosed vulnerabilities affecting NetScaler ADC and NetScaler Gateway appliances. The most critical allows unauthenticated attackers to bypass authentication under certain Gateway and AAA configurations, particularly when security assertion markup language

authentication is enabled. The second is a high-severity memory overflow flaw that can enable denial of service attacks when SIP Application Layer Gateway is enabled in large-scale NAT deployments. Citrix recommends administrators review their configurations, determine whether affected features are in use and upgrade to the latest supported firmware versions. Although neither vulnerability has been observed in active attacks, Citrix noted that previous net-scaler flaws were exploited shortly after disclosure. With more than 23,000 net-scaler instances exposed online, organizations are encouraged to patch promptly to reduce potential risk.

Researchers at Ransom News uncovered more than 50,000 exposed Stripe API keys across public GitHub repositories, GitHub action logs, and misconfigured web servers, demonstrating how leaked credentials can quickly lead to fraud and account compromise. In one case, researchers used an active key to access a merchant's customer data, create a fraudulent payment link, and process a $1 test charge within 17 hours. A recently discovered dataset contained live... API keys for 659 merchant accounts and approximately 35 gigabytes of customer and payment data, although Stripe itself was not compromised. The report highlights that exposed secret keys

often provide full API access, enabling attackers to create charges, issue refunds, and modify webhooks. Researchers recommend rotating exposed keys using restricted API keys, auditing version control history and enabling fraud detection to reduce risk. Researchers at Huntress uncovered a social engineering campaign targeting Black Hat and DefCon attendees, with attackers impersonating a coin desk executive on X to lure victims into opening a malicious Google Doc. The document uses Google Apps script to collect reconnaissance data, including IP addresses, operating system, location, and installed cryptocurrency wallets, before presenting fake errors that prompt users to download malware or run terminal commands.

MacOS victims receive malware resembling atomic MacOS Stealer, while Windows users are targeted with payloads linked to credential theft, remote access, and persistent backdoors. A secondary lure uses a fake Dropbox Docksend site to distribute additional malware, including net support rat. Huntris advises users to treat unsolicited Google Docs like email attachments and avoid documents that request software updates, terminal commands, or security bypasses. Atlassian Splunk and Cisco have released security updates,

addressing more than 400 vulnerabilities across their enterprise products, including numerous critical and high-severity flaws. Atlassian patched roughly 109 unique CVEs affecting products such as Jira, Confluence, and BitBucket, while Splunk fixed more than 150 vulnerabilities across enterprise, SOAR, and related applications. Cisco also resolved 15 vulnerabilities, including multiple critical flaws in crosswork and secure workload that could enable remote code execution, authentication bypass, and other attacks. None of the vendors reported active exploitation, but organizations are urged to apply the updates promptly to reduce risk. Researchers at Threat Fabric have identified Manick, a new Android malware active since at least February,

that combines spyware, banking fraud, and remote control capabilities. Targeting users across Europe, particularly Ukraine, the malware impersonates legitimate apps using transparent overlays to count. PINs, passwords, recovery phrases, and authentication codes, while allowing apps to function normally. After gaining accessibility permissions, it can steal notifications, SMS messages, files, location data, and provide remote access via WebRTC. A notable feature is its fallback data X-filtration method. If a compromised device cannot reach its command and control server, encrypted data is relayed through nearby infected devices using Wi-Fi Direct or Bluetooth, enabling multi-hop transfers.

Users are advised to avoid unofficial apps, limit accessibility permissions, and enable Google Play Protect. CareCloud, a medical data organization, is notifying nearly 3.8 million individuals that personal and health information may have been exposed for, following a March cyber attack on one of its AWS environments. The company said attackers accessed and allegedly exfiltrated data between March 10th and 16th, though no unauthorized activity has been detected since. Potentially affected information includes social security numbers, financial data, medical records,

and insurance information. CareCloud says it strengthened security measures, while multiple law firms are investigating the incident. The breach is currently the third largest health care data breach reported in 26 to the U.S. Department of Health and Human Services. Bit Defender Labs has uncovered Silk Parasite, a cyber espionage campaign assessed with medium confidence as China-linked that has targeted government organizations across Central Asia since late 2025. Researchers identified seven remote access tool families, including, five previously undocumented, designed with modular plug-in-based architectures that minimize detection through in-memory execution and DLL side-loading. Unlike mass-produced AI-generated malware,

Silk Parasite appears to have been professionally engineered, with limited evidence suggesting AI-assisted development rather than AI-written code. The campaign relies on spearfishing documents tailored to regional government agencies and uses trusted services, including Google Drive, for command and control. BitDefender found links to known China Nexus malware families and infrastructure, but stopped short of attributing the operation to a specific threat group. Researchers warn that the campaign demonstrates sophisticated, low-profile tradecraft likely to reappear against other regions and sectors. The Cybersecurity and Infrastructure Security Agency is exploring the use of a contractor to manage enterprise-wide cybersecurity software procurement through a single acquisition process.

A recently issued Sources sought notice marks the beginning of market research for a potential contract supporting approximately $600 million in annual cyber software purchases, with the potential to expand to $6 billion over the contract's lifetime. Working with the General Services Administration's assisted acquisition services, SISA is seeking support for enterprise licensing, software asset management, vendor management, procurement analytics, and strategic acquisition planning. The effort aligns with SISA's transition from the continuous diagnostics and mitigation program's approved product list

to a new cyber product list and technical capability catalog, which will serve as the foundation for future federal cybersecurity software acquisitions. Coming up after the break, my conversation with Chris Wallace from Intruder. We're discussing how AI agents killed the annual pen test and are reshaping exposure management. And AI-powered robots find bananas quite appealing. Stay with us. Chris Wallace is CEO and founder of Intruder.

I caught up with him at Black Hat for this sponsored industry voices segment, discussing why the annual Penn Test is dead and how AI agents are reshaping exposure management. We're continuing our conversations here at Black Hat 2026. And joining me now is Chris Wallace. He is founder and CEO at Intruder. Chris, thanks so much for taking the time for us here today. Thanks having me. So I think there are maybe some organizations who are still out there,

and they're going through their annual calendar, and they're saying, what's time for our annual pen test? We've been doing it this way for years. You're saying that you can't do that anymore. That's not the way it works. I guess when I'm saying you can't. I think some people still do and still will for a while because of the compliance angle,

because that's how compliance frameworks work. they say that you have to do an annual pen test. What we are saying is we don't think it makes sense anymore. And we don't think it's made sense for a long time. It's something that we still do because that's the way it's always been done. But in the last 20 years, everything's changed under the hood that makes it make so little sense anymore. Back in the day, you could do an annual test. They would find some stuff.

And then, yeah, you're going to be secure for six months. Nothing changes. Landscapes are the same. Your systems don't change. Now you have systems changing on database. new code being released multiple times a week if you have applications. The threat landscape is changing on a daily basis. You know, hundreds of thousands of vulnerabilities released every year.

And attackers are weaponising them in hours sometimes. So the idea that you can wait a whole year to check your security is laughable, frankly, isn't it? It's unfortunately this slightly comedic situation that we now find ourselves in, but the compliance isn't catching up. So, you know, we're releasing AI PIN. intration testing to help solve some of those problems. But we kind of anticipate that some people will still do it once a year. And we think that has to change, but it's not going to change overnight.

How do organizations meet both of those needs? Those compliance rules aren't going away, but they understand that they need to do more. So you get two different types of company. And some companies tick the box and they do what they have to do, or whether it's because that's all the budget they can get, is to just tick the box. They have a compliance or a customer requirement and they have to do it

and they've got no more budget for any more security. But then you get other organizations that are a little more proactive and care about security for customer trust or whatever other reason they find the budget and they can do both so they can tick their annual compliance checkboxes but then they can also do things more frequently

and avoid the real problems. How has AI changed how we think of a pen test? in general? It's changing a fair bit, I would say. Firstly, humans are good at certain things and AI is good at other things. So, for example, ingesting a massive code base is not something human can easily do.

My background is penetration testing. You'd get put on a five-day job to test someone's application. And they'd offer to give you a source code, but it's not always that helpful because they can't ingest hundreds of thousands or millions of lines of code and just find weaknesses very quickly. An AI can, you know, an AI can just ingest multiple code bases at once

and go straight to pattern matching and pull out critical weaknesses in minutes. So the speed of the AI being able to do that is completely game-changing. And we've seen results like 65 plus criticals found in a set of applications in four hours. And the human equivalent would have been weeks and weeks to do that same amount of work. So, yeah, the pace that AI can work is incredible. But it doesn't fully replace the human. I think there's stuff that the AI is good at that, you know, finding niche things that humans might have missed. But humans maybe have the edge at following methodologies and being consistent in a way that the AIs are still struggling with.

You know, they're a bit non-deterministic. And they, as we've seen last week with hugging face, sometimes they like to go rogue, right? Well, what does it look like for an organization when they adopt this transition? What's the onboarding and what sort of changes should they expect? Good question. I think what we've seen so far is people adopting this when they have problems to solve. They have an audit requirement. They need to get it done really quickly.

they can't actually go and get a traditional pen test done because that will take weeks and weeks to onboard a supplier, scope it out, get the work done, get the report back. You know, you're talking five, six weeks is a quick timeline. And these people need this done right now. So they've kind of switching to the AI pen test as a way to solve those kind of timing problems, the speed and the timing problems.

So we've not necessarily seen too many people kind of switching over from the old way of doing things just yet. We think we anticipate that's going to happen. but where we've seen the first groundswell is people who are just needing to get this done right now. So they're going less change and more just jumping straight on where they can

because there's a gap. And I think that other people will come it's going to take a bit more time. Are you seeing any acknowledgement on the compliance side that things need to change? Yeah, I think you talked to auditors. Anyone in the industry can look at this situation

and understand that it's not right. It doesn't make sense. But compliance has a way of not changing too quickly, right? There's someone's going to have to come down from the top and say, this whole thing needs to shake up. Frankly, is anyone currently incentivized to do that across any of these schemes? Not really, right? It's just working for a lot of people the old way.

So I can't see it being changed up too soon, but I think it's going to have to change for sure. What does success look like in your mind? When an organization has adopted the kinds of things that you all propose, how does that change their day-to-day operations? More confidence, more trust that they're secure and knowing that they're doing the right security. And I think most security practitioners want to be doing the right thing. They don't want to be following the compliance. It's more like they're forced into the bad way of doing things.

And we've had people come to us to say, I love what you're doing. I love how this can be more continuous, so I can run this all the time. But I'm being forced into this old way of doing things by the compliance requirements. So there's a friction there that can be solved in the future. And I think we'll get there. And there'll be a, there's more Nirvana state for people, right? That they actually know that what they're doing matters and they're getting better security outcomes as a result. So I think, yeah, everyone being a lot happier with their job is probably a nice outcome from this.

It's pretty good. Yeah. stand right now when it comes to the reliability of AI pen testing? What is the state of the art? Yeah, it's a really good question. It's early days with that. And so we're seeing some challenges around the consistency of the things. So they do tend to go rogue. You can do multiple pen tests and sometimes they'll come out with different answers because they spent time on different things, whereas a human would maybe follow a methodology and be a bit more consistent. You wouldn't want to buy a pen test and then get another one done.

And then the second one would find stuff that the first one missed. That would be not a great outcome from a human-led penetration test. And that the AIs have that slight issue when you run them is that they'll go and spend time on something. And then the next time they might spend time on something else. One of the ways to get around that is to give it more time. So it has the time to do a fuller job and look at everything. So it doesn't run out of tokens midway through the tests.

We've tried with testing. giving it a methodology, but we found better results when we don't. So that's one of the areas where we're still working to find the middle ground, because you want the best results you can get. You don't want to tie it down with the methodology and then miss out on the things that it could have found. But at the same time, we need to solve that problem where running two different tests is going to give you two different answers. I have to tip my hat to you.

We're here at Black Hat, and to hear you talk about your aspiration, that we don't have everything figured out. We're working on these hard problems. We're confident we're going to get there, but that's not the type of messaging. I think you hear much on the show floor here. I think it's refreshing. Yeah, thank you.

Yeah, we're consciously trying to do that at the moment because it is such a new area, and I think a lot of customers are still figuring out this stuff as well. I don't think it's ever helped for security vendors to over-promote the stuff that they've got. I think it helps the whole industry to know where things are at. And I come from that background. I'm a pen tester at heart.

And I've always seen, you know, the vendor messaging. And it's like second nature. If you work in security, you have to see the messaging and then start thinking about the truth behind it and what it can do and what it can't do. And then you find out sometimes it can't do what they're saying. I think trust is so important in cybersecurity that vendors have to be truthful and honest. And so, yeah, we're saying we're working on this stuff. It's early days. We're excited about it.

It's incredibly powerful. Some of the stuff we've seen it do is amazing, but we're not going to pretend that it's perfect, right? I think that's the flaw. I think if we explain to customers where it's amazing and where they can get really good value, then they can confidently come on board with us and use it for the right things, whilst not throwing out the kitchen sink or the stuff that is working with humans now, they can keep those deployed but use the AI where it makes the most sense. What are your recommendations then for the folks out there who feel it,

as though they need to be doing a better job with this. Maybe they're a little intimidated by the velocity of change, these new tools. How should they get started? Give us a call, I guess. I mean, it's super easy to get going. You log into the platform. You give us access to your GitLab or your GitHub, you know, code repos.

And you can be up and running in minutes. So if anyone's interested, it's very, very easy to get started with this stuff. And, you know, it doesn't cost the earth either. we're talking like a few thousand dollars for a pen test and see the results see if you like it and yeah what we've seen so far is some really incredible results the stuff that can pick out issues that have survived multiple years of human penetration tests the AI just goes and finds in in seconds so to my point earlier I'm not trying to pretend that this is a perfect technology but the value is incredible like in literally a few thousand dollars and a few hours you can find critical

flaws that you weren't aware of, and I think that's a really important message to be getting out there. What do you think the future looks like? Is it going to be all AI doing the pen testing? Is it a hybrid thing where there's still be a place for humans? Yeah, I think so, yeah. I think humans are always going to have a place, but it might change what they spend their time on. Just as old school automation, you know, like automated scanners came and started doing what humans were doing by hand before, AI is the same. It's going to come and take more load off the human, but I think the human will just be elevated and it'll just be a new level of automation really. I don't think we're going to remove humans from the loop anytime soon.

Well, Chris Wallace is founder and CEO of Intruder. Chris, thanks so much for taking the time for us. Thanks so much. And finally, the future of robotics may arrive with fewer dramatic robot uprisings and more bananas pressed into unexpected service. during a visit to robotics startup Generalist AI, a reporter from Wired witnessed a robotic arm demonstrating an ability to learn on the fly, even improvising by using a banana as a tool,

a small but telling glimpse of how AI-powered machines are becoming more adaptable. Founded by former Google DeepMind and Boston Dynamics researchers, Generalist is taking a different approach to robot training, Instead of relying solely on massive data sets or open-source language models, the company collects high-quality demonstrations using custom camera-equipped grippers worn by human trainers. The result is a proprietary foundation model designed to generalize across tasks rather than failing when something is trivial as the lighting as the lighting changes. Experts say the company's combination of strong engineering and carefully curated training data puts it, among the closest contenders to deploying truly useful robots in real-world commercial environments.

Some robots are still learning to grasp the world. This one apparently started with the produce aisle. And that's the Cyberwire. For links to all of today's stories, check out our daily briefing at thecyberwire.com. Don't forget to check out the Grumpy Old Geeks podcast where I contribute to a regular segment on Jason and Brian's show every week. You can find Grumpy Old Geeks where all the fine. podcasts are listed. 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 Marriott. Maria Vermazas. Our executive producer is Jennifer Ibin. Peter Kilpe is our publisher, and I'm Dave Bittner. Thanks for listening. We'll see you back here tomorrow.