Cleartext – April 25, 2026
Daily cybersecurity briefing for CISOs and security leaders.
Episode Summary
Today's episode covers 18 stories across 6 topic areas, including: A dozen allied agencies say China is building covert hacker networks out of everyday routers; AI Tools Are Helping Mediocre North Korean Hackers Steal Millions; Iran’s cyber threat may be less ‘shock and awe’ than ‘low and slow,’ officials say.
Stories Covered
🌍 Geopolitical
A dozen allied agencies say China is building covert hacker networks out of everyday routers
CyberScoop · Apr 23 · Relevance: █████████░ 9/10
Why it matters to CISOs: This multi-nation advisory signals a fundamental shift in Chinese APT tradecraft toward industrialized botnets from compromised edge devices—CISOs need to map and baseline traffic from all edge devices, especially routers and VPN concentrators.
12 allied agencies issued joint advisory on Chinese government-linked covert networksChina-backed groups now using large-scale botnets built from compromised routers and edge devicesAdvisory includes actionable guidance for organizations to detect compromiseAI Tools Are Helping Mediocre North Korean Hackers Steal Millions
Wired Security · Apr 22 · Relevance: ███████░░░ 7/10
Why it matters to CISOs: AI is democratizing offensive capabilities for nation-state hackers—even less-skilled DPRK operators stole $12M in three months using AI for everything from malware coding to fake websites, lowering the bar for sophisticated attacks.
DPRK hackers used AI for malware development, fake company websites, and social engineeringOne group stole up to $12 million in three monthsAI enables 'mediocre' hackers to punch well above their weightIran’s cyber threat may be less ‘shock and awe’ than ‘low and slow,’ officials say
The Record (Recorded Future) · Apr 24 · Relevance: ██████░░░░ 6/10
Why it matters to CISOs: Reframing Iran's cyber threat as opportunistic and persistent rather than dramatic changes defensive priorities—CISOs in critical infrastructure should focus on detecting stealthy, long-dwell intrusions rather than preparing solely for destructive attacks.
US officials assess Iran's most likely cyber approach is opportunistic intrusions dressed up to look biggerState-sponsored and hacktivist groups refining attacks on energy, water, and critical infrastructureSeparate reporting shows AI-generated malware (ZionSiphon) targeting Israeli water systems, though assessed as ineffective📡 Macro Trends
Mozilla: Anthropic's Mythos found 271 security vulnerabilities in Firefox 150
Ars Technica Security · Apr 21 · Relevance: █████████░ 9/10
Why it matters to CISOs: Mythos finding 271 bugs in a mature codebase signals AI-driven vulnerability discovery is now operationally real—CISOs must consider both the defensive acceleration this enables and the offensive implications if adversaries gain similar capabilities.
Anthropic's Mythos AI model found 271 security vulnerabilities in Firefox 150Mozilla CTO says model is 'every bit as capable' as world's best security researchersRaises concerns about offensive use and collapsing exploit windowsUnchecked AI Agents Cause Cybersecurity Incidents at Two Thirds of Firms
Infosecurity Magazine · Apr 21 · Relevance: ████████░░ 8/10
Why it matters to CISOs: Two-thirds of firms reporting AI agent-related incidents validates the urgency for CISOs to establish governance frameworks for autonomous AI agents before they become the next major attack surface.
Cloud Security Alliance report finds 2/3 of firms experienced AI agent-related cybersecurity incidentsIssues include data exposure, operational disruption, and financial losses85% of enterprises running AI agent pilots but only 5% trust them enough for productionIn a first, a ransomware family is confirmed to be quantum-safe
Ars Technica Security · Apr 23 · Relevance: ███████░░░ 7/10
Why it matters to CISOs: Ransomware adopting post-quantum cryptography signals that recovery via future cryptanalytic breakthroughs is off the table—CISOs must double down on prevention, backup integrity, and resilience rather than hoping for future decryption capabilities.
First confirmed ransomware family using post-quantum cryptographyNo practical benefit today but ensures encrypted data cannot be decrypted even with future quantum computersSignals ransomware operators are future-proofing their operations🔓 Data Breach
Vercel attack fallout expands to more customers and third-party systems
CyberScoop · Apr 23 · Relevance: ████████░░ 8/10
Why it matters to CISOs: The Vercel breach—originating from a single employee's OAuth grant to a third-party AI tool—illustrates why CISOs must govern OAuth token grants and third-party AI tool integrations as first-class attack surfaces.
Breach originated from compromised third-party AI tool (Context AI) via stolen OAuth tokensVercel found additional compromised customer accounts after expanding investigationSignificant downstream risk to customers and third-party systems; Mandiant engaged‘Scattered Spider’ Member ‘Tylerb’ Pleads Guilty
Krebs on Security · Apr 21 · Relevance: ███████░░░ 7/10
Why it matters to CISOs: The Scattered Spider guilty plea reinforces that social engineering and SMS phishing remain the most effective initial access vectors—CISOs should ensure help desk and employee authentication procedures are hardened against these techniques.
Tyler Robert Buchanan, 24, pleaded guilty to wire fraud conspiracy and aggravated identity theftPart of Scattered Spider group that hacked 12+ major tech companies via SMS phishingGroup stole tens of millions in cryptocurrencyFormer Ransomware Negotiator Pleads Guilty to Working For BlackCat Cyber Gang
Infosecurity Magazine · Apr 22 · Relevance: ███████░░░ 7/10
Why it matters to CISOs: This insider threat case is a wake-up call for CISOs using third-party ransomware negotiation services—vetting, separation of duties, and oversight of IR vendors must be rigorous to prevent conflicts of interest.
Ransomware negotiator Angelo Martino pleaded guilty to secretly working with BlackCat operatorsHelped maximize ransom amounts while employed as a legitimate negotiatorWorked with BlackCat starting April 2023 across five different victimsUK Biobank Data Breach: Health Data of 500,000 Listed for Sale in China
Infosecurity Magazine · Apr 24 · Relevance: ███████░░░ 7/10
Why it matters to CISOs: Half a million health research records appearing on Chinese e-commerce platforms underscores the geopolitical dimension of health data theft—CISOs in healthcare and research organizations must treat biomedical data as a national security asset.
Health records of 500,000 UK Biobank volunteers listed for sale on Chinese e-commerce platformsUK government minister confirmed the breachRecords were removed after discovery but exposure window unknownNew Checkmarx supply-chain breach affects KICS analysis tool
BleepingComputer · Apr 23 · Relevance: ███████░░░ 7/10
Why it matters to CISOs: A security vendor's own analysis tool being compromised via Docker images and VS Code extensions is a trust-shattering event—CISOs must verify integrity of security tooling itself and implement supply chain controls even for trusted vendors.
Hackers compromised Docker images and VS Code extensions for Checkmarx KICS security analysis toolMalicious tags overwrote existing versions to harvest data from developer environmentsSecurity tooling itself became the attack vectorADT confirms data breach after ShinyHunters leak threat
BleepingComputer · Apr 24 · Relevance: ██████░░░░ 6/10
Why it matters to CISOs: ADT's breach by ShinyHunters highlights the ongoing risk from extortion groups targeting consumer-facing companies with large customer datasets—CISOs in similar verticals should review data minimization practices and extortion response playbooks.
ADT confirmed breach after ShinyHunters threatened to leak stolen dataCustomer data stolen; ransom demandedHome security company with millions of customer accounts⚖️ Governance & Policy
CISA director pick Sean Plankey withdraws his nomination
CyberScoop · Apr 22 · Relevance: ████████░░ 8/10
Why it matters to CISOs: CISA's leadership vacuum deepens at a time of escalating threats—CISOs relying on federal coordination and guidance should plan for continued uncertainty in federal cyber leadership and policy direction.
Sean Plankey withdrew after waiting over a year for confirmationCISA has been operating under chaotic temporary leadershipWithdrawal leaves the nation's primary cyber agency without a confirmed leaderHouse Republicans roll out national privacy bill
CyberScoop · Apr 22 · Relevance: ███████░░░ 7/10
Why it matters to CISOs: A federal privacy law would reshape data governance obligations across the enterprise—CISOs should evaluate how their current privacy programs align with Virginia/Kentucky-style frameworks this bill is modeled on.
House Republicans released federal privacy legislationBill modeled on Virginia and Kentucky state privacy lawsLacks bipartisan support, creating uncertainty about passageLatest spy power reauthorization bill leaves critics unimpressed
CyberScoop · Apr 24 · Relevance: ██████░░░░ 6/10
Why it matters to CISOs: Section 702 reauthorization directly affects how intelligence agencies can compel enterprise cooperation on surveillance—CISOs at cloud and telecom providers should track this for potential new compliance obligations.
April 30 deadline looming for Section 702 FISA powers extensionNew legislation drawing fire from both left and rightReauthorization debate has implications for enterprise data handling obligations🚀 Startup Ecosystem
Why Cisco Is Eyeing Buy of Non-Human Identity Startup Astrix
BankInfoSecurity · Apr 25 · Relevance: ███████░░░ 7/10
Why it matters to CISOs: Cisco's potential $250-350M acquisition of Astrix validates non-human identity as a critical security category—CISOs should evaluate their own NHI posture given that machine identities now vastly outnumber human ones.
Cisco reportedly in talks to acquire non-human identity startup Astrix Security for $250-350MRepresents 25%+ premium to Astrix's last $200M valuationWould expand Cisco's identity security beyond authentication and ITDR🚨 Critical Vulnerability
CISA Hunts for Cisco Backdoor Spotted on Federal Network
BankInfoSecurity · Apr 25 · Relevance: █████████░ 9/10
Why it matters to CISOs: The Firestarter backdoor surviving reboots, upgrades, and standard remediation on Cisco firewalls means CISOs must reassess their edge device integrity verification processes and potentially plan for hardware-level inspections.
CISA issued emergency directive over 'Firestarter' backdoor on Cisco Firepower/ASA devicesMalware survives reboots, upgrades, and standard remediation processesJoint US-UK advisory; discovered on a federal civilian agency network compromised since September 2025Three AI coding agents leaked secrets through a single prompt injection. One vendor's system card predicted it
VentureBeat Security · Apr 21 · Relevance: ████████░░ 8/10
Why it matters to CISOs: Prompt injection in AI coding agents affecting Claude, Gemini, and Copilot demonstrates that AI-assisted development workflows are now a concrete attack vector—CISOs must enforce strict controls on AI agent permissions in CI/CD pipelines.
Researchers demonstrated prompt injection via GitHub PR titles that made Claude, Gemini CLI, and Copilot Agent leak API keysAttack requires no external infrastructure—just a malicious PR titleAffects workflows using pull_request_target trigger, which most AI agent integrations requireFurther Reading
🌍 A dozen allied agencies say China is building covert hacker networks out of everyday routers — CyberScoop🌍 AI Tools Are Helping Mediocre North Korean Hackers Steal Millions — Wired Security🌍 Iran’s cyber threat may be less ‘shock and awe’ than ‘low and slow,’ officials say — The Record (Recorded Future)📡 Mozilla: Anthropic's Mythos found 271 security vulnerabilities in Firefox 150 — Ars Technica Security📡 Unchecked AI Agents Cause Cybersecurity Incidents at Two Thirds of Firms — Infosecurity Magazine📡 In a first, a ransomware family is confirmed to be quantum-safe — Ars Technica Security🔓 Vercel attack fallout expands to more customers and third-party systems — CyberScoop🔓 ‘Scattered Spider’ Member ‘Tylerb’ Pleads Guilty — Krebs on Security🔓 Former Ransomware Negotiator Pleads Guilty to Working For BlackCat Cyber Gang — Infosecurity Magazine🔓 UK Biobank Data Breach: Health Data of 500,000 Listed for Sale in China — Infosecurity Magazine🔓 New Checkmarx supply-chain breach affects KICS analysis tool — BleepingComputer🔓 ADT confirms data breach after ShinyHunters leak threat — BleepingComputer⚖️ CISA director pick Sean Plankey withdraws his nomination — CyberScoop⚖️ House Republicans roll out national privacy bill — CyberScoop⚖️ Latest spy power reauthorization bill leaves critics unimpressed — CyberScoop🚀 Why Cisco Is Eyeing Buy of Non-Human Identity Startup Astrix — BankInfoSecurity🚨 CISA Hunts for Cisco Backdoor Spotted on Federal Network — BankInfoSecurity🚨 Three AI coding agents leaked secrets through a single prompt injection. One vendor's system card predicted it — VentureBeat SecurityFull Transcript
Click to expand full episode transcript
Jordan: Every once in a while you get a week where the threat landscape doesn't just evolve — it accelerates. This was that week. Chinese APT groups industrializing botnets out of your edge devices. A confirmed backdoor on Cisco firewalls that laughs at your remediation playbook. AI finding two hundred and seventy-one zero-days in Firefox in what I can only assume was an afternoon. If you took a few days off this week, welcome back — you have some reading to do.
Alex: You're listening to Cleartext. I'm Alex Chen, with me as always is Jordan Reeves, and this is our Saturday Week in Review. If the daily episodes piled up in your queue, no judgment — here's what actually mattered this week and what it means for your program going into Monday. We've got four big themes to work through. Nation-state tradecraft took a significant step forward, and we have the joint advisories to prove it. The AI threat surface expanded in three distinct and uncomfortable directions simultaneously. The breach ledger had some entries that should shape how you think about third-party risk and your own IR vendors. And on the governance side, the leadership vacuum at CISA got worse. Let's get into it.
Jordan: So the geopolitical story of the week — and I'd argue the story of the quarter — is that twelve allied agencies put out a joint advisory on Chinese APT groups building covert networks out of compromised routers and edge devices. Twelve nations. That's not a routine threat bulletin, that's a coordinated diplomatic signal dressed up as technical guidance. What they're describing is a fundamental shift in Chinese offensive tradecraft. These groups have moved away from targeted, bespoke intrusion infrastructure toward industrialized botnets built from the devices sitting at the edge of your network right now. Your SOHO routers, your VPN concentrators, your firewalls. The tradecraft advantage here is real — when your malicious traffic originates from a legitimate IP in someone's home or small business network, attribution becomes genuinely hard, and detection requires a baseline you probably don't have.
Alex: And I want CISOs to hear that last part clearly. Do you have a baseline for what normal looks like on your edge devices? Not your endpoints — your routers, your concentrators, your network edge. For most organizations the honest answer is no, and that's precisely the gap these groups are exploiting. The advisory does include actionable detection guidance, which is relatively unusual for a document like this. Your team should pull that and work through it this week.
Jordan: The second China-adjacent story that connects here is the UK Biobank breach. Half a million health research records — genomic data, health histories — appeared for sale on Chinese e-commerce platforms. A UK government minister confirmed it. And I think CISOs in healthcare and research need to sit with that for a second. This isn't just a GDPR problem. Biomedical data at scale is a national security asset. The adversary understands that even if your legal team is still framing it as a privacy compliance issue.
Alex: Alongside the China picture, we got some important context on Iran this week that I actually found useful as a framing correction. US officials are pushing back on the idea that Iran's cyber posture is primarily about shock-and-awe attacks. The assessment is more nuanced — Iran is operating low and slow, opportunistic intrusions that are designed to look bigger than they are. There's separate reporting about AI-generated malware targeting Israeli water infrastructure, assessed as ineffective, but the direction of travel matters. The implication for CISOs, particularly in energy, water, and critical infrastructure — you're not primarily preparing for a dramatic destructive event. You're hunting for stealthy, long-dwell intrusions that have been in your environment for months.
Jordan: Now let's talk about the AI theme, because it ran through almost every story this week in different ways. And I want to separate three distinct vectors here because they're easy to conflate. First is AI as an offensive force multiplier for adversaries. The Wired reporting on North Korean hackers this week was genuinely clarifying. We've spent a lot of energy worrying about nation-state actors with elite technical capabilities. The DPRK story flips that — one group of operators that analysts describe as mediocre stole up to twelve million dollars in three months using AI for malware development, fake company websites, social engineering scripts, the works. AI is not just making good hackers better. It's making average hackers dangerous. That changes your threat model.
Alex: The second AI vector is AI as a vulnerability discovery engine, and this one is frankly the most consequential for the long term. Anthropic's Mythos model found two hundred and seventy-one security vulnerabilities in Firefox 150. Mozilla's own CTO said the model is every bit as capable as the world's best security researchers. Let me put that in board-level language. The window between vulnerability existence and exploitation just collapsed. If a well-resourced adversary has access to similar capability — and we should assume they either do or will soon — the notion that you have thirty days to patch a disclosed vulnerability becomes a fiction. Your patching cadence, your compensating controls, your detection posture — all of it needs to be rethought against a world where vulnerability discovery is automated and fast.
Jordan: The third AI vector this week was internal — AI agents as a new attack surface inside the enterprise. Two stories converged on this. Cloud Security Alliance data showed two thirds of firms have already experienced cybersecurity incidents related to AI agents — data exposure, operational disruption, financial loss. At the same time, researchers at Johns Hopkins demonstrated prompt injection attacks through GitHub PR titles that caused Claude, Gemini CLI, and GitHub Copilot to leak their own API keys. The attack required no external infrastructure. Just a malicious pull request title. If your developers are running AI coding agents in CI/CD pipelines — and eighty-five percent of enterprises are piloting this — you have an active attack surface that most security teams are not governing.
Alex: And I want to add the Checkmarx story here because it connects. Hackers compromised Docker images and VS Code extensions for Checkmarx's own KICS security analysis tool. A security vendor's tooling became the attack vector. When you add that to the AI agent prompt injection research, you're looking at a developer environment where the tools themselves cannot be fully trusted. That's a supply chain problem with implications for how you govern your software development lifecycle end to end.
Jordan: On breaches — and there were several worth your attention — the Vercel incident is the one I'd spend the most time on. The origin point was a single employee's OAuth grant to a third-party AI tool called Context AI. The breach expanded to additional customer accounts during the investigation, Mandiant got called in, and the downstream exposure is still being scoped. This is the story you use when your board asks why you care about OAuth token hygiene. One OAuth grant to one AI productivity tool opened a breach that is affecting Vercel's customers and their customers. Map your OAuth grants. Govern third-party AI tool integrations like the attack surface they are.
Alex: Two guilty pleas this week that are worth flagging. First, a member of Scattered Spider — Tyler Buchanan, twenty-four years old — pleaded guilty to wire fraud and aggravated identity theft. His group hacked twelve major tech companies using SMS phishing. There is nothing technically sophisticated about what Scattered Spider does. It is social engineering, it is help desk manipulation, it is exploiting the gap between your security policies and how your people actually behave under pressure. If you haven't hardened your help desk authentication procedures and employee identity verification recently, this is your reminder. The second plea was more disturbing. A ransomware negotiator named Angelo Martino admitted he was secretly working for BlackCat while nominally representing victims in negotiations. He was helping maximize ransom amounts against the organizations that hired him. I don't want to overstate the frequency risk here, but the implication is clear — your IR vendor relationships need oversight, vetting, and separation of duties. Trust but verify applies to the people you call when things go wrong.
Jordan: There's also the Firestarter backdoor story that broke Friday and I want to make sure it doesn't get lost in the weekend news cycle. CISA issued an emergency directive over a backdoor on Cisco Firepower and ASA devices that survives reboots, upgrades, and standard remediation. A joint US-UK advisory. The backdoor was discovered on a federal civilian agency network that had been compromised since September 2025. Seven months. If you have Cisco edge devices — and most of you do — your current remediation playbook is insufficient for this threat. You need to pull the CISA guidance and think about what hardware-level integrity verification actually looks like in your environment.
Alex: Quickly on the governance front — CISA's nominated director Sean Plankey withdrew his nomination this week after waiting more than a year for Senate confirmation. I don't want to editorialize too much on the politics but the operational implication is real. CISA is the federal coordination layer for threat intelligence sharing, emergency directives, and incident response support. It has been operating in chaos. If any part of your security program leans on federal guidance or coordination — and it probably should — you need contingency posture for the possibility that coordination remains degraded for the foreseeable future. Build more peer relationships. Lean harder on ISACs. Don't assume the federal backstop is reliable right now.
Jordan: On the privacy legislation front — House Republicans released a federal privacy bill modeled on Virginia and Kentucky state frameworks. No bipartisan support, uncertain path to passage. Our take — don't restructure your data governance around this bill. But do use the moment to benchmark your current program against where federal legislation is likely to land eventually. The direction is set even if the timing isn't.
Alex: And the market note — Cisco reportedly in talks to acquire non-human identity startup Astrix Security for somewhere between two-fifty and three-fifty million. For CISOs this is a category validation signal. Machine identities now vastly outnumber human ones in most enterprise environments and most programs are not governing them with anything like the rigor applied to human identity. If you don't have a clear answer to the question of what non-human identities exist in your environment and what access they have, that's a gap that deserves priority this quarter.
Jordan: So what was this week? If I had to name it — this was the week the attack surface became simultaneously wider and faster. Wider because edge devices, AI agents, developer tooling, OAuth grants, and IR vendors all showed up as active attack vectors in real incidents. Faster because AI-driven vulnerability discovery and AI-enabled adversaries are compressing the timelines we've historically built our programs around.
Alex: The through-line I'd bring to your board is this. For the better part of a decade the security conversation was about complexity — too many tools, too much data, too much noise. This week was a preview of a different problem. Adequate defenses are getting harder to define because the offense is accelerating faster than most programs can track. That's not a counsel of despair — it's an argument for ruthless prioritization. Edge device integrity. AI agent governance. OAuth hygiene. Help desk authentication. Those are your four action items going into the week.
Jordan: Get those Cisco advisories in front of your team Monday morning. Don't let the weekend bury them.
Alex: That's the week. Thank you for spending part of your Saturday with us. Cleartext returns Monday with daily coverage. If this episode was useful, share it with a peer who needed the catch-up. We'll see you next week.
Cleartext is an automated daily podcast for CISOs and security leaders. Generated 2026-04-25.
Sources are pulled from: CyberScoop, The Record, SecurityWeek, Krebs on Security, Dark Reading, Cybersecurity Dive, BleepingComputer, Wired, Ars Technica, TechCrunch, Help Net Security, VentureBeat, Risky Business News, The Hacker News, CISA, and BankInfoSecurity.