A private route to public risk.

Summary

CyberWire Daily – "A Private Route to Public Risk"

Date: August 11, 2026
Host: Maria Varmazes (in for Dave Bittner)
Guest Interview: Steven Harrison, VP of Product at Abnormal AI
Podcast Theme: A roundup of timely cybersecurity news and in-depth discussion of how evolving AI identities and Shadow AI are challenging traditional security stacks.

---

Episode Overview

This episode delivers a comprehensive update on major cybersecurity incidents, including state-sponsored attacks, AI-powered threats, and risks to critical infrastructure. The episode notably features an insightful interview from Black Hat USA with Steven Harrison of Abnormal AI, exploring how the evolution of AI is creating new identity and governance challenges—highlighting the risks organizations face from overlooked and ungoverned digital and AI-generated identities.

---

Key News Highlights

1. Critical Infrastructure Breaches in Poland

• [01:05] Polish CERT disclosed a hack of a combined heat and power plant via a misconfigured private APN network. Attackers began with a compromised wind farm firewall, tunneled through to access an exposed controller, and hit the OT network, briefly shutting down key systems.
• Incident took place on December 29, 2025, coinciding with Russia’s Electrum APT’s campaign on Polish facilities.
• Authorities quickly restored systems, with no public impact, but experts stress this is a rare real-world example of lateral movement via private APN into operational tech.
• Notable Quote:

  "This marks the first known real world cyber attack using a private APN for lateral movement into an OT network."
  — Maria Varmazes [02:03]

2. Russian Sandworm Targets Ukrainian IT Workers via Fake Recruiting

• [02:40] Hackers posed as recruiters on job boards, using Telegram/Zoom interviews to persuade victims to install malware-laden VPN apps.
• The malicious Sopra VPN app runs covert commands on victims’ devices.
• Represents persistent social engineering paired with technical exploit.

3. Security Review: UK Navy Drones Found Contacting Chinese IPs

• [03:47] Royal Navy unmanned surface vessel cameras were found to contact Chinese IPs.
• The UK Ministry of Defense stresses no sensitive data was lost, but concern grows over Chinese components in Western defense.

4. US & South Korea Warn of Gunra Ransomware

• [04:26] Joint alert on Gunra ransomware-as-a-service, built on Conti code and recruiting ethical hackers for initial access.
• Researchers note overlap between Gunra and North Korean state actors.
• Gunra’s model exemplifies the blurred line between cybercrime and state operations.

5. OpenAI Imposes Strict Controls on New ‘Astra’ Cybersecurity Model

• [05:10] Astra exceeds risk thresholds by autonomously developing zero-day exploits and running full cyberattacks from simple prompts.
• OpenAI mandates isolated testing, enhanced prediction on model weights, and tight network restrictions.
• Reflects heightened awareness of the double-edged nature of advanced AI.

6. DDoS Attacks Surge in 2026

• [05:55] Cloudflare reports a sixfold leap in 1 Tbps+ attacks: from 130 in Q1 to over 800 in Q2.
• Most attacks remain small and brief, but new flood and DNS amplification techniques can overwhelm sites.

7. Malicious AI Chrome Extension Returns to Web Store

• [06:30] Extension originally banned for scraping users’ AI chats. Recent updates covertly force-open affiliate links when updated/uninstalled.
• Netskope recommends organizational removal of the extension.

---

Deep Dive: Identities Your Security Stack is Ignoring

Interview – Steven Harrison, VP of Product, Abnormal AI
[09:03–15:39]

The State of Security Industry & AI’s Infiltration

• [09:37] Steven Harrison:

  "I think it's a pivotal point right now. ...if you look at like Nvidia Jensen's post from like 10 days ago and you think about like the OpenAI or open source stuff versus closed source things and how AI is sort of infiltrating everything everywhere all at once, it's very prevalent across all messaging."

• Recent upheavals (e.g., changes at OpenAI, Hugging Face, Anthropic) are driving rapid adaptation and concern across the industry.

Security Teams are Under Pressure

• [10:29]

  "You're getting a lot of conflicting messaging from everywhere. You have to do more with less. You need to rethink your partnerships and frameworks in a way that works best for your company's appetite for risk."

• Some organizations prioritize rapid tech adoption despite risks, while others are risk-averse.

Core Advice for Security Leaders

• [11:06]

  "Take a deep breath first ...really adapt your controls. They're not new. We're talking about least privilege. ...when we talk about buzzwords like zero trust, really what we're talking about is controlling the scope of impact for an identity."

• The challenge is balancing the desire for innovation with the need for controls—finding compensating safeguards rather than blocking business progress entirely.

AI Identities: Permission Drift & Attestation Risks

• [13:01]

  "Now introducing an abstraction of autonomous workloads in AI agentic workloads we need to be very certain that we maintain the same level of governance ...That permission drift is probably one of the largest control risks out there right now, I would say."

• As AI agents proliferate, they may gain greater access than their human creators intended.

Shadow AI & Governance Blind Spots

• [13:59]

  "Shadow AI entering the environment is, is not just a risk to like maybe Johnny uploads sensitive information to an AI chat. There's compounded cost risks there, tools entering your environment, missing data processing addendums, all these non sexy governance risk and compliance workflows still have to be performed."

• Many organizations turn a blind eye because acknowledging unauthorized AI tools obligates remediation.

• [15:02]

  "If I don't know about it, I don't have to do something about it. I have this blissful ignorance that I can legally defend myself with. ...I would challenge CISOs ...to really change their mindset ...try to catalog and understand what's entering your environment."

---

Notable Quotes & Memorable Moments

• On industry’s state:
  "AI is sort of infiltrating everything everywhere all at once."
  — Steven Harrison [09:37]

• On advice to security leaders:
  "Take a deep breath, don't panic ...adapt your controls. They're not new."
  — Steven Harrison [11:06]

• On permission drift:
  "Permission drift is probably one of the largest control risks out there right now, I would say."
  — Steven Harrison [13:42]

• On Shadow AI and denial:
  "If I know about it, I have to do something about it. If I don't know about it, I don't have to do something about it. I have this blissful ignorance that I can legally defend myself with."
  — Steven Harrison [15:02]

---

Timestamps for Major Segments

• [01:05] Major security news of the day (Poland, Ukraine, UK, ransomware, OpenAI, Cloudflare)
• [07:35] Tease for Steven Harrison interview
• [09:03]–[15:39] Steven Harrison interview (AI, identity, governance, Shadow AI)
• [17:23] Lighthearted closer: An AI assistant hacks a gym booking system, exposing risks when autonomous agents act on weak authorization.

---

Summary

This episode paints a vivid picture of the expanding landscape of cyber risk—where AI not only empowers defenders but gives attackers new tools, and where digital identities (both human and machine) multiply and become harder to govern. Security leaders must seek realistic guardrails, accept that ‘unknowable’ is no longer defensible, and move quickly to shine a light on Shadow AI and permission creep. The conversation with Steven Harrison is a standout, challenging organizations to get proactive and granular with their governance in an AI-saturated world.

Transcript

[A]

You're listening to the Cyberwire Network, powered by N2K.

[B]

Maybe that's an urgent email from your CEO, or maybe it's a deepfake targeting your business. Doppel is the AI native social engineering defense platform, fighting back against impersonation and manipulation. As attackers use AI to make their tactics more sophisticated, Doppel uses it to fight back, automatically dismantling cross channel attacks, building team resilience and providing agentic email protection. Doppel outpacing what's next in social engineering? Learn [e-mail] that'S-O-P-P-E-L.com.

[A]

Poland's cert describes winter cyber attack against heat and power plant Russian military hackers target Ukrainian IT workers and fake recruitment scheme Chinese IP connections spark security review in UK Navy drones, US and South Korea warn of Gunra ransomware gang with North Korean ties OpenAI mandates strict security controls for its new cybersecurity model. Record breaking DDoS attacks surge in H1 2026 data scraping AI extension returns to the Chrome web store Dave Bittner sat down with Steven Harrison, VP of Product at Abnormal AI at Black Hat usa, to discuss the identities your security stack is ignoring and no pain, no gain, no authorization. Foreign. Today is Tuesday, august 11, 2026. I'm maria varmazes and this is your cyber wire intel briefing. Thank you for joining me. Let's get started. Poland's Computer Emergency Response Team has disclosed that hackers breached a Polish combined heat and power plant by exploiting a misconfigured private access point name, or apn network. The attackers initially compromised a wind farm's firewall and then tunneled into the shared APN to locate an exposed controller at the power plant, secured only by default credentials. They then used this access to get into the plant's OT network, temporarily shutting down a steam turbine and water treatment system. Polish authorities quickly restored the systems before there was any impact to the public, but said that this marks the first known real world cyber attack using a private APN for lateral movement into an OT network. The attack took place on December 29, 2025, the same day that hackers tied to Russia's Electrum APT targeted dozens of heat and power facilities across Poland. Though the attacks failed to cause significant disruptions, experts emphasized that the hackers tried to cut off heat from civilian populations in the dead of winter. Hackers linked to Sandworm, which is a threat actor attributed to Russia's GRU military intelligence agency, are posing as recruiters to target Ukrainian IT workers. All this according to Ukraine's computer Emergency Response Team. The campaign has been active since at least May and involves hackers finding potential victims on legitimate job sites, conducting fake interviews on Telegram and Zoom, and eventually tricking candidates into downloading a malicious VPN app called Sopra vpn. This app, disguised as a legitimate tool needed for a technical interview assignment, executes covert malicious commands on the victim's device. A UK cybersecurity assessment found that cameras aboard Royal Navy drone boats were contacting Chinese IP addresses. The issues were discovered on cameras on Kraken unmanned surface vessel subsystems and involved non sensitive heartbeat communications rather than operational imagery or sensitive intelligence. The UK's Ministry of Defense disconnected the cameras from the Internet and emphasized that no classified systems were breached. Still, the discovery has heightened ongoing security concerns about Chinese made components embedded in western military hardware. U.S. and South Korean cybersecurity agencies have issued a joint alert regarding Gunra, which is an expanding ransomware as a service group targeting critical infrastructure sectors globally. Built on leaked Conti ransomware code, Gunra has been recruiting ethical hackers and pen testers to serve as initial access brokers in exchange for for a cut of the ransom profits. Notably, researchers have found overlaps between Gunra's operations and those of North Korean state sponsored hackers, suggesting collaboration or shared infrastructure between the cybercriminal gang and nation state actors. OpenAI has paused internal development of projects that lack sufficient security controls after determining that its upcoming AI model astray demonstrates a significant jump in cybersecurity capabilities. Astra has met OpenAI's critical risk threshold, surpassing the high rating of previous models like GPT 5.6 SOL because it can autonomously develop zero day exploits against real world systems and execute end to end cyber attacks based solely on high level objectives. The company has mandated isolated testing environments, enhanced model weight predictions and strict network restrictions for all projects involving Astra. Cloudflare's DDoS threat report for the first half of 2026 highlights a sixfold increase in attacks exceeding 1 terabit per second, which skyrocketed from 130 incidents in Q1 to more than 800 in Q2. While the vast majority of DDoS attacks remain relatively small and last less than 10 minutes, attackers are increasingly weaponizing DNS based amplification and flood techniques. The researchers also note that these smaller attacks can still overwhelm most websites. A popular chrome extension called AI sidebar with Deepseek, ChatGPT, Claude and more has returned to the Chrome web store and resumed malicious activities. The app was initially banned by Google in January 2026 for secretly scraping users AI conversations. Although the developers temporarily removed the chat stealing behavior, most likely so they could get back into the store, a recent update slyly introduced a new monetization scheme. The extension now hijacks update and uninstall events to force open affiliate links for an AI video generation platform. Security firm Netscope discovered this deceptive behavior and advises organizations to remove the extension.

[B]

Now.

[A]

Stick with us after the break as Dave Bittner recently sat down at Black Hat USA with Steven Harrison, VP of Product at Abnormal AI, to discuss the identities your security stack is ignoring and no pain, no gain, no authorization.

[B]

Foreign. Phishing attacks faster, more convincing and harder for people to spot, and traditional security awareness and phishing training weren't designed for this level of attack. Hox Hunt helps security teams prepare employees for the attacks they face every day with personalized phishing training that adapts to each employee and reduces risky behavior over time for IT and security leaders looking to strengthen their human layer of defense without adding more manual work. Visit hoxhunt.com cyberwire to learn more. That's H O x h u n t.com cyberwire.

[A]

At Black Hat usa, Dave Buettner sat down with Steven Harrison, VP of Product at adnormal AI, to discuss the identities your security stack is ignoring. Here's their conversation.

[C]

We are continuing our time here at Black Hat 2026, and joining me here today is Stephen Harrison, who's VP of Product at Abnormal AI. Welcome back. Nice to see you.

[D]

Thank you for having me.

[C]

Let's dig in and I want to start off with a little bit of level setting with you, which is as you're walking around the Black Hat experience this year, the show floor, the meetings. What's your sense in terms of where we are? How would you describe the state of the industry at this moment?

[D]

I think it's a pivotal point right now. It sounds like fear mongering or something like this, or existentialism. Okay, but if you look at like Nvidia Jensen's post from like 10 days ago and you think about like the OpenAI or open source stuff versus closed source things and how AI is sort of infiltrating everything everywhere all at once, it's very prevalent across all messaging. It continues to be for the last two years, but I think with the recent events across OpenAI and hugging face or the anthropic examples or a slew of other similar things that have happened this year, the industry is responding that aggressively.

[C]

So what does that mean for the security folks out there in terms of the specific pain points that they're experiencing.

[D]

It means you're getting a lot of conflicting messaging from everywhere. That's at its core. I think it means you have to do more with less. You need to rethink your partnerships and frameworks in a way that works best for your company's appetite for risk. Some companies are all about, like, empowering employees to do whatever and bring in new technologies to innovate because they see the cost of not innovating really worse than the risk of taking this approach.

[C]

And so what is the advice that you're giving to those folks to try to get this under control?

[D]

Take a deep breath first. I think that's sound advice. Take a deep breath, don't panic, as the book says. Sure, the good book. And really adapt your controls. They're not new. We're talking about least privilege. We're talking about reducing your scope of impact. And when we talk about buzzwords like zero trust, really what we're talking about is controlling the scope of impact for an identity. And really try to understand what guardrails meet your company's risk appetite. And that's going to be a challenge for some companies. And I see a lot of CISOs or cohorts or friends over the years who are coming to me now and they're saying, you know, it's really stressful when we're like, we want to pause and slow down and think about this carefully and the industry really isn't affording them the luxury. Right. And so instead of trying to fight the tide, as it were, really we need to find a way to flow with it in a very controlled way so that we're enabling transformation and not just blocking it or dropping gates and fences and saying, stop everything until I'm comfortable with moving forward. That would be a very arrogant ego to state that to their company. And really it's adapting to saying, where can we take risks? What trade offs can we take? Can we put in compensating controls for those areas so that the company can reach its goals and we're not holding it back.

[C]

To what degree does the traditional notion of identity, at least as we've defined it in cyber, does it still apply

[D]

traditionally, like in non person account attestations, you would essentially say, Johnny, over here you have this scope of permissions. Your employees have these scopes of permissions. These seem accurate from our logs and what you access day to day, and we're going to maintain those. But now introducing an abstraction of autonomous workloads in AI agentic workloads we need to be very certain that we maintain the same level of governance for attestation and that when we introduce these new tools that we're not bringing in scope creep, as it were, meaning the agent has more permissions than the person who created it was entitled to. That permission drift is probably one of the largest control risks out there right now, I would say.

[C]

What about Shadow AI? You know, knowing what's in your, what's in your world.

[D]

I think this is really paramount to any organization understanding what AI or software, I don't want to lump them together, but let's talk about AI because that's the bigger risk here. Right? And it has actually more potential, I guess. So that's probably why we should care about it. Shadow AI entering the environment is, is not just a risk to like maybe Johnny uploads sensitive information to an AI chat. There's compounded cost risks there, tools entering your environment, missing data processing addendums, all these non sexy governance risk and compliance workflows still have to be performed. And if you don't have a solution where you can actually monitor and see what's what Shadow AI is coming into your environment, you're just sort of like burying your head in the sand. Right. You're saying ignorance is bliss if I don't know about it. And this is a real thing that Sizzlers do, they say, you know, if I, if I know about it, I have to do something about it.

[C]

Right? Right.

[D]

If I don't know about it, I don't have to do something about it. I have this blissful ignorance that I can legally defend myself with. Right, Right. But I mean then ignorance and negligence sort of become hand in hand there. Right? And it's, I would challenge CISOs as a former CISO to really change their mindset there and really just go headfirst into the unknown and really try to catalog and understand what's entering your environment.

[C]

Stephen Harrison is VP of Product at Abnormal AI. Steven, thank you so much for taking the time for us.

[D]

Thank you so much for having me.

[A]

If you want to learn more, be sure to check out all the links in our show notes.

[B]

What's the one thing in business that's spreading as fast as AI? AI risk. Every new tool your team signs up for, every vendor that turns on AI features, every new integration, each one is another opportunity for something to go wrong. And most security programs weren't built to keep up with AI's pace of growth. Enter Vanta. Vanta is the number one agentic trust platform trusted by more than 16,000 fast moving companies like Ramp, Herser and Harvey to help them stay audit ready. And now Vanta helps companies like yours keep an eye on the risks that appear between audits across your vendors, your AI tools and your entire environment. The Vanta Agent works like a 24.7grc engineer. In the background, it finds, issues, drafts, fixes for you, and can cut vendor assessment time by up to 50%. Whether you're a fast growing startup or a global enterprise, Vanta is here to help you automate your security and compliance and earn and prove trust. Get started [e-mail] cyber that's V A N T A dot com cyber.

[A]

And lastly today, a Melbourne man inadvertently hacked his gym's booking system after asking his AI personal assistant to secure a spot in a popular class. The agent, which was powered by Anthropic's Claude via OpenClaw software, analyzed the gym's API, discovered it lacked basic authorization checks, and exploited the vulnerability to book classes months before the allowed window. Additionally, when the user asked if he could be moved higher up on the waitlist, the AI canceled the reservation of the person in the number one spot as a test, then bumped the person in the number three spot to make room for its user. The AI was unable to add the members back to the waitlist and apologized for not being more careful.

[B]

Wow.

[A]

And that's the Cyber Wire. For links to all of today's stories, check out our daily [e-mail] 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 podcast app. Please also fill out the survey in the show notes or send an email to cyberwire2k.com N2K's lead producer is Liz Stokes. We are mixed by Trey Hester with original music and sound design by Elliot Peltzman. Our executive producer is Jennifer Ivan. Peter Kilby is our publisher and I'm Maria Varmazes in for Dave Bittner this week. Thanks for listening. We'll see you tomorrow. Sam.