CyberWire Daily - What the Flock?

Starting point is 00:00:00 You're listening to the Cyberwire Network, powered by N2K. Maybe that's an urgent email from your CEO, or maybe it's a deep fake targeting 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, automatically dismantling cross-channel attacks, building team resilience, and providing agentic email protection. Dopple, outpacing what's next in social engineering. Learn more at doppel.com. That's do pp-p-el.com. The G7 and Syssa prepare for the quantum threat.

Starting point is 00:01:08 Nightmare Eclipse drops a crowd strike zero day. The White House's offensive hacking plan raises legal questions. Sisa offers a playbook for community. communicating through cyber incidents. Open AI puts a billion dollars behind AI-powered defense. Researchers uncover a serious post-GrescuL flaw. Google patches an exploited Chrome Zero Day. Broadcom fixes VMware vulnerabilities.

Starting point is 00:01:32 Attackers target a WordPress plug-in flaw. Lawmakers tell license plate surveillance cameras to flock off. Our guest is Kevin Gostchalk, founder and CEO of Arcos Labs, discussing his new book AfterBots, which questions the old assumption that automated traffic is inherently malicious. And camouflage for the algorithmic age. It's Friday, September 4th, 2026. I'm Dave Bittner, and this is your Cyberwire Intel briefing.

Starting point is 00:02:09 Thanks for joining us here today. It is great as always to have you with us. The G7 and SISA are urging governments and businesses to start preparing now for the cybersecurity risks posed by quantum computing. While the timeline for sufficiently powerful quantum machines remains uncertain, advances could eventually allow attackers to break widely used encryption. The threat isn't entirely in the future. Attackers can steal encrypted data today and hold on to it until quantum computers can decrypt it, the so-called Harvest Now decrypt later strategy. That puts long-lived secrets,

Starting point is 00:03:02 including government records, personal information, and corporate intellectual property at particular risk. The advisory recommends identifying sensitive systems and critical assets, prioritizing them for migration to post-quant cryptography, and incorporating quantum-resistant technology into routine upgrades. Starting early, officials say, should make the transition cheaper and less disruptive. The guidance follows similar efforts in the U.K., which is called for organizations to complete their migration to quantum-resistant cryptography by 2035. The security researcher who goes by the name Nightmare Eclipse has expanded their zero-day campaign beyond Microsoft, publishing vulnerabilities affecting several endpoint security products.

Starting point is 00:03:54 The latest, dubbed Falcon Flank, is a privilege-escalation flaw that allegedly abuses Crowdstrike Falcon's Microsoft Office macro removal feature. Crowdstrike says it's investigating and recommends customers disable the affected policy setting while retaining other office malware protections. Security researcher Kevin Beaumont says he confirmed Falcon Flank Works, along with several other recently released exploits from Nightmare Eclipse. Those include hard breacher, an elevation of privilege vulnerability affecting Kaspersky endpoint security, and Pretty Prague, an avast flaw that can reportedly provide system privileges. Gen Digital says it's developing a patch for the avast issue.

Starting point is 00:04:43 Nightmare Eclipse has also published Green Section and NVIDIA memory corruption Zero Day that Beaumont says can crash affected systems. Kaspersky and Invidia did not respond to requests. for comment. The White House Initiative allowing vetted private companies to conduct cyber operations against foreign criminal groups could strip those firms of legal protections they currently use for defensive cybersecurity work, experts warn. President Trump's August memorandum puts participating companies under federal control with justice and homeland security officials approving operations.

Starting point is 00:05:23 Former DOJ cybercrime official Leonard Bay. Bailey said that government agent's status could complicate protections under laws governing information sharing, communications interception, and defensive measures. Experts also flagged risks beyond U.S. law. Companies could violate foreign hacking statutes, face retaliation from criminal groups with government ties, and damage their standing in international markets. Threat intelligence executive Alex Orleans also warned that vendors may overestimate their ability to accurately attribute malicious infrastructure. The risks could be especially significant for smaller, inexperienced contractors.

Starting point is 00:06:05 With agencies still developing the program's rules, experts say the legal, financial, and geopolitical consequences need considerably more scrutiny. SISA, the FBI and cybersecurity agencies from Australia, Canada, New Zealand, and the U.K., have issued guidance urging service providers to communicate more clearly during major IT and operational technology outages. The agencies say effective communication can be as important as technical remediation in limiting disruption and speculation. Organizations should prepare communications plans in advance, establish clear roles and backup channels, and tailor messages for technical teams, executives, customers, regulators, and the public.

Starting point is 00:06:55 During an incident, providers should quickly explain what's affected, what's known about the cause, and what remains uncertain, while avoiding speculation and PR language. The guidance also recommends maintaining a single source for continuous timestamped updates, providing actionable technical information, coordinating with legal and law enforcement partners, and eventually explaining the root cause and planned security improvements. OpenAI has pledged $1 billion to subsidize access to its daybreak cybersecurity models for essential services, beginning in the United States and later expanding to partner countries. The daybreak for Frontline Defenders Initiative will help organizations in sectors, including water, electricity, local government, banking, and non-profits integrate OpenAI's models into existing security tools and workflows. OpenAI says daybreak can help resource-constrained defenders analyze suspicious activity,

Starting point is 00:08:01 review legacy code, identify vulnerabilities, prioritize risks, and develop fixes. A pilot with the multi-state information sharing and analysis center, the MSISAC, will provide daybreak access, training, and hands-on support to an initial group of U.S. public sector and water system defenders. The commitment follows an industry warning, that defenders have a narrowing window to use frontier AI before AI-enabled attacks pose greater risks to critical public services. Cybersecurity firm Cyera has disclosed a severe PostGresQL vulnerability,

Starting point is 00:08:41 affecting releases dating back to 2014. Dubbed PostGres Shell, the flaw can allow attackers with replication privileges to execute code and escalate to database super-executive. user. The vulnerability stems from insufficient authorization in PostgreSQL's logical decoding functionality. CYERA found that attackers can supply a file system path as a plugin name, causing PostgresQL to load malicious code with the privileges of the server process. From there, an attacker could access databases, execute operating system commands, steal sensitive files, and establish persistent back doors. Sayera says multiple versions of PostgresQL are affected, fixes are available, and organizations should patch promptly and audit accounts with replication privileges.

Starting point is 00:09:38 Google has updated Chrome to patch an actively exploited zero-day vulnerability in its V8 JavaScript and WebAssembly engine. The high-severity type confusion flaw could potentially allow remote code execution through malicious JavaScript on a a crafted web page. Google has withheld exploitation details while users apply the fix. The update also addresses 11 other vulnerabilities, including nine high-severity memory safety flaws. Updated Chrome versions are rolling out for Windows, MacOS, and Linux. Broadcom has patched two serious VMware workstation and fusion vulnerabilities that could

Starting point is 00:10:21 allow attackers with administrative privileges inside a virtual machine to execute code on the host. One of the vulnerabilities rated critical at 9.3 is an integer overflow affecting systems using the VMX net 3 adapter. The other is a high-severity stack-based buffer overflow. Broadcom says no workarounds are available and recommends updating promptly. attackers are actively exploiting a critical vulnerability in the Elementor Pro WordPress plugin to upload web shells and execute arbitrary commands. The vulnerability affects multiple versions and stems from faulty validation of file upload arrays in Elementor forms. WordFence says exploitation began August 19th, the same day Elementor released their latest version,

Starting point is 00:11:17 which includes the fix. and they've blocked nearly 200,000 attempts. Exploitation requires a published Elementor Pro Form widget with a file upload field. Attackers can bypass validation to upload malicious PHP files, which are then accessible for remote command execution. Administrators should immediately upgrade and inspect the plug-ins form upload directory for PHP files, which WordFence says are a strong indicator of compromise. Representatives Thomas Massey and Eric Berluson have introduced the Flock Off Act,

Starting point is 00:11:57 which would prohibit federal funding for automated license plate readers and biometric surveillance cameras. The bill wouldn't ban flock safety cameras outright, but federal agencies would have to remove federally funded systems, while state and local recipients would have 180 days to stop operating them. toll collection systems would be exempt. The bipartisan group of seven sponsors argues that flocks expanding license plate reader network enables mass surveillance of law-abiding Americans. Flock cameras record information including license plates, vehicle characteristics, time,

Starting point is 00:12:37 and direction of travel. Flock says its technology helps law enforcement solve crimes and locate missing people and stolen vehicles. The proposal comes amid growing local resistance to flock systems, with nearly 150 communities reportedly canceling contracts, rejecting deployments or deactivating cameras over privacy and oversight concerns. Coming up after the break, my conversation with Kevin Gostchalk from Arcos Labs. We're discussing his new book, AfterBots, and camouflage for the algorithmic age.

Starting point is 00:13:22 Today's cybercriminals aren't just launching attacks. They're building businesses around them. They have subscription models. They have a marketplace. They have affiliate program. If you want to do referrals, you can get credits. They make it really easy. It really looks like a legitimate SaaS product that somebody might use.

Starting point is 00:13:53 I sat down with Mike Britton, CIO at Abnormal AI, to explore how AI is lowering the barrier to cybercrime and what security leaders need to change in response. Here are full conversation at explore.thecyberwire.com slash abnormal AI. Kevin Gosschalk is founder and CEO of Arcos Labs. His new book, AfterBots, questions the old assumption that automated traffic is inherently malicious. So I've been in the bot detection space. I've been running this company Arcos for just around 10 years now.

Starting point is 00:14:40 And it's always been the traditional, hey, if bot traffic is coming to our website, that's a bad thing. We don't like that. That really changed late last year to this idea of, you know, bot traffic is now kind of becoming the standard way of how we want to access information. You know, of course, due to all these AI tools like chat GPT and Claude. And on the back of that, there's a lot of new questions we need to ask from the practitioner and detection side, around, okay, is the bot still the malicious kind of old, or is this the consumer use good kind of bot, which obviously are now kind of being called agents, or is this malicious actors using this more modern technology?

Starting point is 00:15:26 So it's really kind of reshaping this whole landscape, and that was the genesis for this book. Can we go through maybe a basic question here, which is, at what point is a bot no longer a bot and it becomes an agent. You know, from a raw technology standpoint, they are virtually the same thing in terms of how it would interact with the workflow. It's autonomous in nature. And historically, that's what we've Christianed bots as.

Starting point is 00:15:58 But there are some nuances to, I think, how we would categorize the old category of bot traffic, which is a little bit more simplistic, you know, doing co-request or basic HTTP request to something that now has the ability to kind of control your operating system, control your phone, control your browser, and reason and think thanks to these LLM harnesses like OpenClaw or Claude Co-work or these kind of technologies. And it's more that modern category that we're kind of calling an agent, even though under the hood, you know, technically it's still doing similar things that I bought used to do. Right.

Starting point is 00:16:37 You know, the book kind of replaces that question, is this a bot with the notion of who is this agent? What's it trying to do? And is the platform willing to authorize that action? It seems to me like we have to start thinking about intent maybe more than we did in the past. Yeah, it's a lot of the detection technologies that we've used historically for things like fraud and security purposes. the same kind of ones that we will use to answer those kind of questions. But it needs to now be a more nuanced question, which is kind of like there's multiple layers like you mentioned.

Starting point is 00:17:17 Like, A, it's first the question of, is it an agent? And then is the agent identifying that it's an agent? Like does it pass like Web Bot Orth API handshake? So you know for a fact that there's an agent? Or is it something more akin to like perplexity where it pretends to be Chrome and it doesn't tell you that's an agent? And then you've got another category, which is bad actors using this technology, generally through means like open claw to do autonomous things.

Starting point is 00:17:46 And, you know, all of those look slightly different from a technology standpoint. And none of those really answer the question of should I allow it. So there's another detection layer, which is this intent concept, which is something like, hey, this agent's creating, you know, multiple accounts in a short time span. clearly that intent is probably like promotional fraud or something versus hey it's logging into my bank account and it's paying my rent and that's something that I do every week and it's doing it at the same cadence that I'm doing it and that's you know that's probably an intent that's okay and then you've now also got authorization which is do I even want agents to be doing these actions

Starting point is 00:18:28 like do I if I'm a bank do I want agents to sign up on behalf of consumers maybe I do Do I want agents transferring money? Maybe I don't. So then you've also kind of got that concept of per workflow. Like what is my internal policy and what am I going to allow from an authorization standpoint? One of the things that really caught my eye was this notion of economic deterrence. Can you unpack that for us? Yeah, so this is when we're talking about how do we deal with threat actors that are targeting and committing fraud

Starting point is 00:19:03 and trying to extract from our ecosystems and our customers' money. So this is like scams and account compromise and promotional fraud, credit card fraud, these kind of things. Ultimately, fraud and cybercrime is a business. They are doing this to make money. And the persistent theme that I've seen over the decade, I've been doing this and everything I've learned from all the amazing companies we have the opportunity to work with is if you can make the cost of

Starting point is 00:19:33 them committing the act higher than the profit that they're getting back from doing so, surprise, surprise, they stop. They'll go do something else. A really simple example, if I sign up for a credit card at some of the largest retailers, they'll give me an incentive for signing up, some kind of promotional bonus. Maybe it's like a $10 gift card or a $50 gift card if I'm approved. And that's generally what these bad actors are targeting, is if my cost is lower than $50 to get approved for a credit card, I can pocket that difference. But if suddenly the platform changes the gift card pay out based on the risk of the sign-up, so if I look like a high-risk user and I only get a $5 gift card versus the $50, maybe now I'm no longer

Starting point is 00:20:17 profiting from that act. Those kind of things will deter users. That premise persists even with an agenetic world where we will now start seeing certain kinds of fraud that really weren't profitable because the cost of having a human in the loop was too high. Now you can kind of remove the human from committing the fraud and you can have an agent fleet, maybe go and open, I don't know, thousand cards an hour. But if you can introduce detection mechanisms

Starting point is 00:20:48 that can A, glean out that these things are autonomous agents and B, start forcing them back to human operations, in the loop again, you maintain that cost advantage against them. So there are, I would say, strategies and technology that will allow us to continue to raise that cost and effort, even for agent harnesses that are trying to do the reverse for threat actors, which is, dramatically decrease the cost of committing fraud. This is a persisting concept that, you know, has existed since almost the beginning of time when it comes to, like, warfare. If the reward is worth as lower than the effort, then of course you're not going to do it, right?

Starting point is 00:21:28 Is it fair to assume that some of these agents are going to recognize what you're up to and try to circumvent it to try to make themselves look like they're up to something other than what they actually are? Yeah, that's the two, from a fraud landscape and cyber standpoint, absolutely. Anything that they're doing that gets them caught in detection rules, are they going to be learning from that and storing that state to memory and then trying to figure out how to reverse engineer those detection rules. The one advantage we always have of the threat actors is they're trying to do things to profit versus your typical customers are just trying to use your

Starting point is 00:22:07 platform in a normal way. So there's got to be some concept of them extracting money out of the ecosystem. And generally, that's a very anomalous behavior. So just the pure reasoning of how they make money. Generally, even if it's an agent or a human or anyone trying to defraud you in these platforms, there is something still to model in that behavior that's unique to threat actors. What are the recommendations that you hope people come away from having read the book? I think one of the big things is first and foremost understanding that this is not some, you know, all-seeing superpower. This is something that you can build a technology framework for. You can build an authorization policy matrix for.

Starting point is 00:22:50 It is something that, you know, once you understand how the technology works and acts, it's very comprehensible to see a world where we can allow good agents and good customers using this technology, as well as continue to deter and mitigate fraudulent threat actors. We don't have to pick one or the other, even though the technology kind of defaults initially to being what appears to be more of a security threat just because of the world looking at bots is bad historically. That's definitely changing now. I'm curious, as you were taking on this task of writing this book,

Starting point is 00:23:26 to what degree did you consider its shelf life in such a rapidly changing space? Was that something that was top of mind for you? You know, it's funny. I gave the book to one of the Frontier Labs, CSOs at Black Hat this year, and I have a page towards the front of the book

Starting point is 00:23:45 that touches on that concept that So this is a very fast-moving space. I think the good news is the underlying technology of how this technology interacts with the Internet is going to persist. I think the frameworks and some of the standards that we're kind of experimenting with, you know, for example, WebBot-Orth may come and go.

Starting point is 00:24:07 But I think the underlying premise and all of that, which it talks to, should be good for quite some time. But, oh, yes, definitely keenly aware that even what I wrote back then is probably already somewhat getting dated. That's why we have to release new additions. There you go. Well, Kevin, I think I have everything I need for our story here. Is there anything I missed, anything I haven't asked you that you think it's important to share?

Starting point is 00:24:34 You know, I think one thing that's going to change the agentic traffic shape of things quite dramatically is, you know, today, most people using agents to go and do consumer things like, you know, booking hotels or booking cars or doing searches, it's really regulated or relegated more to the technology kind of savvy people. Like, what are the early adopters, I would say? But I think very soon the internet traffic is going to shift heavily towards agents going and doing things on your behalf. Once we see one of these technology, you know, the large tech companies like at Amazon or at Google or Microsoft, I think I could probably see that Google will be first to do it,

Starting point is 00:25:22 where in the Android operating system, currently today you see a screen of apps when you turn your phone on. I think very soon, when you turn your phone on, you'll see a prompt screen and you'll tell it what you want it to do for you, and it will have agents go and do that work. They'll go to the various websites, they'll go source the various bits of information, and they'll make it actionable,

Starting point is 00:25:42 and your UI will change from apps to a prompt, and it'll get the results, and it will bring them to you. And I think once that change happens, they could also, of course, do the same thing with the Chrome browser instead of surfacing search results. Just do the work for you, like what Claude and OpenAI do. I think once that happens, and that could happen really any month, the vast majority of internet traffic will shift agents.

Starting point is 00:26:06 So it's coming very soon, I think. That's Kevin Gossach, founder and CEO of, Arcos Labs. And finally, Berlin artist Simon Wechert has found a fashionable way to protest AI surveillance, become invisible to the algorithm. His digital camouflage uses colorful patterns designed to confuse Yolo, a popular family of object recognition models. Hold the fabric in front of a person, and the camera's reassuringly confident person label

Starting point is 00:26:57 suddenly has second thoughts. Weckert developed the project after Berlin police deployed the city's first object recognition surveillance cameras capable of identifying people and potentially flagging behaviors such as loitering or someone lying down. By repeatedly modifying patterns and testing Yolo's confidence, he produced designs that can prevent the algorithm from recognizing a person. There is a catch. Police haven't disclosed what technology.

Starting point is 00:27:27 their cameras use, so Weckert can't say his camouflage works against them. And as recognition systems evolve, his patterns will need updates. Surveillance-resistant fashion apparently also has seasonal collections. And that's the Cyberwire. For links to all of today's stories, check out our daily briefing at thecyberwire.com. We will not be publishing this coming Monday in recognition of the federal holiday. We will instead have a special edition in your Cyberwire feed, featuring myself Maria Ramazas and Brandon Carf. We're discussing concerns about China's presence inside U.S. telecommunications systems. Do check it out.

Starting point is 00:28:19 And hello, Maria Vermazes here. On Sunday's T-Minace face cyber briefing. We're talking about attacks on timing with Andy Davis, who is the Global Research Director at NCC Group. That is on Sunday on T-minus. Don't miss it. Be sure to check out this weekend's Research Saturday and my conversation with Ismail Valenzuela,

Starting point is 00:28:40 Vice President of Labs, threat research and intelligence at Arctic Wolf. We're discussing their work tracking Anubis. That's Research Saturday. Check it out. 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.

Starting point is 00:28:59 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 Pelksman. Our contributing host is Maria Vermazas. Our executive producer is Jennifer Ibin. Peter Kilpy is our publisher and I'm Dave Bittner. Thanks for listening. We'll see you back here next week.

CyberWire Daily - What the Flock?

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.