Coverage: Last 24 hours

Today’s Highlights

Rapid acceleration of AI-driven code and exploitation is compounding remediation backlogs and raising organizational exposure, while the highest-risk AI super-users are quietly embedding opaque tools deep in business workflows. Key security themes: AI-fueled attack surface expansion, remediation debt from fast code cycles, concentration of AI risk in super-users, operational impacts of deepfake proliferation, national security adoption of AI-enabled threat detection, and looming regulatory changes pressuring datacenter operations.

Defensive Actions

• Automate inventory and risk assessment processes for open-source packages imported by AI coding tools, and regularly prune unnecessary dependencies.
• Deploy AI code scanning and SBOM integration to ensure every production deployment is analyzed for vulnerabilities and hidden dependencies.
• Identify and closely monitor AI tool usage among top 5% of internal power users, with special scrutiny of unauthorized integrations and workflows.
• Test EDR and intrusion detection controls against AI-enabled attacker TTPs, including rootkit and memory-resident threats targeting Linux and Windows infrastructure.
• Prepare incident response playbooks for deepfake harassment and reputational attacks, coordinating with legal and communications teams.
• Update analyst procedures to bypass AI-generated search summaries and rely on direct review of full threat intelligence sources.
• Collaborate with school administration to craft clear policies governing AI use, balancing enablement with visibility and monitoring.
• Track key regulatory proposals related to AI and datacenter operations, and proactively map out required infrastructure and compliance adjustments.
• Roll out organization-wide deepfake awareness, reporting, and remediation training tailored to staff and student safety.

Table of Contents

1. Shipping More AI Code Than You Can Secure? Watch How to Control Remediation Debt
2. ⚡ Weekly Recap: AI-Powered PLC Attacks, GitLab Attacks, Stripe Key Leaks and More
3. The Outsized Shadow: Why 5% of AI Users Are Your Biggest Security Risk
4. UAT-10147 Uses AI to Scale Server Attacks, Deploys SPECTRE With EDR Bypass and Linux Rootkit
5. Tell us: do you think AI has made Google search better or worse?
6. UK to use Ukraine battlefield data to train AI to protect sensitive sites
7. Albanese seeks to quell datacentre disquiet as climate expert warns ‘we’ve got one shot to get the rules right’
8. How to encourage smarter AI use in the classroom
9. The Download: kids outlearning AI, and space travel agents
10. Kids outlearn AI, and we still don’t know why
11. Black Box: episode 2 – The hunt for ClothOff, the deepfake porn app – podcast
12. They Dedicated Their Lives to Teaching. Then the Deepfakes Started

Top Stories

Shipping More AI Code Than You Can Secure? Watch How to Control Remediation Debt

Source: The Hacker News | Risk: HIGH | Impacted: DevOps pipelines using AI coding tools, SaaS product engineering teams, Security teams relying on manual code review

Summary: If your developers are using AI coding tools, you are probably already seeing the upside: faster development, more code, and less time spent on routine work. The harder part is what comes after. AI can also introduce open-source packages at a pace your security team was never built to handle. More dependencies mean more vulnerabilities to review, more remediation work,

Why it matters: Accelerated development pipelines driven by AI-generated code and mass open-source adoption are creating remediation bottlenecks and mounting organizational technical debt, directly increasing the window of exploitability for real attackers.

Practitioner Perspective

Security and engineering teams face swelling backlogs as AI-assisted development hastens code delivery and brings in more third-party packages than legacy processes can safely review. The sheer rate of package churn means unpatched vulnerabilities rapidly outpace available remediation capacity, eroding assumptions about how quickly exposure can be reduced. This calls for not just improved code scanning, but deliberate pruning of unnecessary dependencies and, crucially, organizational investment in modernizing SBOM management and supplier risk processes. Without this, critical flaws may linger for months below the radar, especially in systems built with AI-generated code. Lead the conversation on technical debt by pushing for automation in dependency hygiene and meaningful escalation paths for vulnerabilities identified via AI code review.

Recommended Actions

• Automate inventory and risk assessment for new open-source packages added via AI coding assistants
• Integrate AI code scanning and SBOM tools, ensure every deployment incorporates automated dependency analysis

⚡ Weekly Recap: AI-Powered PLC Attacks, GitLab Attacks, Stripe Key Leaks and More

Source: The Hacker News | Risk: MEDIUM | Impacted: Organizations relying on PLC/ICS, Teams self-hosting GitLab, Environments handling sensitive payment APIs

Summary: A package gets installed. A login prompt opens. A box sits exposed to the internet. Nothing looks unusual yet. That’s roughly the mood this week. Trusted tools turn hostile, old weak spots get fresh attention, AI makes exploit work cheaper, and researchers keep finding attacks that sound harder than they actually are. Plenty to clean up. Here’s the short version.

Why it matters: Attackers are weaponizing automation and trusted software supply chains to compromise targets quickly and at scale, making it increasingly difficult for defenders to rely on traditional detection or response playbooks.

Practitioner Perspective

Recent activity spotlights a rising tempo of attacks, AI-powered exploits against PLCs, supply chain threat exposure (GitLab, Stripe keys), and the hostile repurposing of legitimate tooling. Each of these increases operational stress, especially in environments without robust anomaly detection or automated credential hygiene. Expect more blended attacks that straddle IT and OT, using AI-generated payloads to leapfrog traditional defenses. If your defenses assume slow-moving campaigns or single-vendor threats, it’s time to re-examine playbooks and cross-disciplinary incident response. Never underestimate the velocity advantage AI now gives both commodity and advanced threat actors.

Recommended Actions

• Monitor for AI-trained attack TTPs targeting PLC and OT equipment, baseline and alert on anomalous logic updates
• Audit GitLab self-hosted instances for known supply chain exposures and credential leaks

The Outsized Shadow: Why 5% of AI Users Are Your Biggest Security Risk

Source: The Hacker News | Risk: HIGH | Impacted: Business units leveraging AI for workflow automation, IT teams with limited shadow IT visibility

Summary: Big security risks come in small packages. While enterprise security teams focus on policing the proliferation of employees using ChatGPT and Claude for quick drafting tasks, a more urgent threat is posed by a handful of AI super-adopters who are quietly hardcoding unvetted tools into critical business operations. According to new research published by Akamai, the top 5% of enterprise

Why it matters: A small number of AI ‘power users’ embedding unauthorized automation and scripts can quietly create systemic risk that far exceeds the perceived threat from broader, lightweight end-user adoption.

Practitioner Perspective

Security teams tend to focus on mass AI adoption across staff, but research shows the greatest risk is caused by super-users who hardwire unapproved AI tools directly into business-critical processes. This places opaque dependencies and unvetted models deep in production, sidestepping normal controls, classic shadow IT at a scale that security leaders often underestimate. Proactive AI governance should target these pockets of risk, not just casual chatbot use. Focus on mapping and risk-assessing workflows operated by high-leverage users, and actively monitor for unsanctioned API and automation activity. Failing to do so leaves the organization exposed to unmitigated third-party risk and potential data leakage.

Recommended Actions

• Identify and audit AI tool usage among top 5% of users by API volume or automation scripts
• Establish inventory and approval processes for AI integrations tied into mission-critical workflows

UAT-10147 Uses AI to Scale Server Attacks, Deploys SPECTRE With EDR Bypass and Linux Rootkit

Source: The Hacker News | Risk: HIGH | Impacted: Internet-facing Linux/Windows server fleets, Organizations in education, media, tech, and gaming

Summary: Cybersecurity researchers have disclosed details of a Chinese-speaking cybercrime group dubbed UAT-10147 that’s targeting Windows and Linux web servers globally across the education, media, technology, and gaming sectors. The vast majority of the targets are located in Brazil, Bolivia, China, Canada, and Vietnam. Details of the threat activity came to light following the discovery of an open

Why it matters: Sophisticated attackers are leveraging AI to automate infrastructure compromise at scale and evade standard EDR controls, raising the risk that defenders will miss new TTPs involving AI-generated malware and Linux rootkits.

Practitioner Perspective

The UAT-10147 group exemplifies how adversaries use AI to rapidly discover, exploit, and persist on Linux and Windows servers, including with EDR bypass and SPECTRE-linked rootkits. Organizations running web-facing infrastructure, particularly in high-traffic sectors or with exposure to rapidly developed AI features, are at particular risk. The tactics described show attackers mixing AI automation with evasive malware deployment, reducing detection time and operational friction. Security teams must not assume their existing EDR baselines apply: threat emulation or purple teaming against AI-flavored TTPs is rapidly becoming essential. Prioritize threat hunting for memory-resident tools and rootkit artifacts, don’t wait for EDR alerts.

Recommended Actions

• Hunt for SPECTRE-based rootkits and memory-resident implants on Linux endpoints
• Test EDR coverage against AI-generated attack TTPs, including bypass attempts documented in UAT-10147 reports

Tell us: do you think AI has made Google search better or worse?

Source: The Guardian | Risk: MEDIUM | Impacted: Threat intelligence analysts, SOC teams conducting open-source research

Summary: As people grow used to AI chatbots, we’d like to hear your views about Google’s search engine Google has put artificial intelligence at the front and center of its search bar. The most-visited site on the internet still shows the same list of links to users, but they have to scroll past a summarized response from an AI chatbot, a

Why it matters: AI-generated search summaries can filter out relevant threat intelligence, potentially limiting analysts’ ability to quickly access primary sources and reducing situational awareness.

Practitioner Perspective

Defenders increasingly rely on web search for open-source threat intelligence, including for emerging vulnerabilities and attack techniques. With AI summarization at the front of platforms like Google, there’s higher risk of critical context being omitted or misrepresented, especially for technical queries. Analysts may need to adjust workflows to scroll past AI sections and verify information directly against authoritative sources. Training teams to spot hallucinations or misleading summarization artifacts becomes necessary. The most reliable intelligence still comes from reading full advisories and original disclosures.

Recommended Actions

• Update analyst workflows to bypass AI search summaries and prioritize manual review of primary sources
• Document procedures for verifying intelligence context missed by AI-generated results in Google Search

UK to use Ukraine battlefield data to train AI to protect sensitive sites

Source: The Guardian | Risk: MEDIUM | Impacted: Operators of critical infrastructure, Physical security teams at sensitive sites, Organizations piloting government AI platforms

Summary: London and Kyiv in deal to help stop protesters and hostile states targeting military bases and critical infrastructure AI models trained on Ukrainian battlefield data will be used to stop protesters and foreign states targeting UK defence sites, railways and energy plants under a deal struck between London and Kyiv. Private companies will also be given access to the vast

Why it matters: Military-grade AI models bring new detection capabilities but also introduce unique classification and privacy risks when adapted to civilian critical infrastructure contexts.

Practitioner Perspective

As governments begin deploying battlefield-trained AI to defend infrastructure from state and activist threats, defenders will inherit systems designed for very different adversary and privacy models. This may improve response to coordinated physical and cyber incidents but could also raise concerns about surveillance or overbroad alerting. Private sector participants must be ready to audit model logic and advocate for proportional controls in sensitive environments. Maintain observability on how AI-derived alerts influence physical security and cyber incident response. The convergence of military and civilian applications means any new anomaly-detection AI should be vetted for unintended impacts on operations and civil liberties.

Recommended Actions

• Engage in review of AI detection models sourced from military applications before deployment on civilian infrastructure
• Establish logging and monitoring to track false positives and alert fatigue in AI-driven physical security use

Albanese seeks to quell datacentre disquiet as climate expert warns ‘we’ve got one shot to get the rules right’

Source: The Guardian | Risk: MEDIUM | Impacted: Operators of hyperscale and enterprise data centers, Security compliance teams, Facilities management organizations

Summary: Prime minister will use national cabinet meeting to assuage premiers over new AI law as Aemo forecasts seven-fold rise in datacentre power use Follow our Australia news live blog for latest updates Get our breaking news email, free app or daily news podcast Anthony Albanese will seek to use Wednesday’s high-stakes talks with premiers to quell growing unhappiness about national

Why it matters: Regulatory uncertainty and increasing energy demands linked to AI growth create operational instability for data centers, forcing CISOs to anticipate rapid changes to compliance and infrastructure planning.

Practitioner Perspective

With policymakers debating how to regulate AI’s impact on national critical infrastructure, particularly data center power consumption, security leaders are left in a reactive posture. This cloud of uncertainty exposes organizations to policy whiplash: sudden compliance deadlines, reporting requirements, or even mandatory architectural changes. Data center operators should press for early clarity from legislators and prepare for significant adjustments in resiliency, reporting, and physical security expectations. Monitor how new AI laws are shaping sector-specific regulatory frameworks. The most resilient orgs will preemptively map AI-driven energy and security scenarios and advocate for clear, actionable requirements.

Recommended Actions

• Track pending AI-related regulatory proposals affecting datacenter power use and operational disclosure
• Engage legal and facilities leadership to map infrastructure changes required by new AI governance requirements

Emerging Signals

Black Box: episode 2 – The hunt for ClothOff, the deepfake porn app – podcast

Source: The Guardian | Risk: HIGH | Impacted: Employees and students subject to public exposure, Legal and HR departments, Incident response teams handling harassment claims

Summary: Revisited: Guardian journalist Michael Safi looks into the world of artificial intelligence, exploring the dangers and promises it holds for society Today in Focus is on a summer break and will be back with new episodes from 1 September. In the meantime, we are bringing you season one of Black Box, before the launch of season two in early September.

Why it matters: The ease of deepfake production raises the risk of targeted reputation attacks, harassment, and blackmail campaigns that security teams will be asked to triage even if current controls were not built for this threat class.

Practitioner Perspective

While much coverage of deepfakes focuses on celebrities, attackers are already weaponizing this technology to target regular staff, students, and executives. Security and legal teams will see growing pressure to respond to incidents where deepfake content is part of targeted harassment or extortion. Defenders must pre-plan how to respond, including rapid legal escalation, digital evidence collection, and public relations coordination. Relying on technical controls alone is insufficient, incident response need to extend into policy, awareness, and victim support. The probability of being forced to remediate deepfake incidents even outside the IT department is now uncomfortably high.

Recommended Actions

• Develop incident response playbooks for deepfake harassment, including legal escalation contacts
• Coordinate with communications and HR for unified response to reputational attacks fueled by AI content

They Dedicated Their Lives to Teaching. Then the Deepfakes Started

Source: The Verge AI | Risk: HIGH | Impacted: Teachers and school staff, K-12 and university HR teams, School IT and incident response

Summary: The deepfake epidemic in schools is affecting more than students. Four teachers tell WIRED about becoming targets of sexualized, AI-generated content, and how difficult it was to find accountability.

Why it matters: Targeted deepfake campaigns now threaten not only high-profile individuals but also regular employees, expanding the type of reputational and wellbeing risk organizations must be prepared to respond to at scale.

Practitioner Perspective

The emergence of deepfake-based targeting of educators is a harbinger for broader employee risk, where doctored content can create real damage with few existing playbooks or support structures. Security and HR must align to address these incidents, including rapid takedown of malicious content and support for affected individuals. Because deepfakes can circulate far beyond their technical point of origin, defenders cannot treat this solely as a cybersecurity issue; cross-department escalation paths are now required. Training staff to recognize and report deepfake abuse is mandatory, not optional. Every organization needs a unified stance on deepfake response, waiting until after an incident is irresponsible.

Recommended Actions

• Roll out deepfake awareness and reporting training to staff, emphasizing abuse and reputational harm vectors
• Work with social networks and hosting platforms to automate the takedown of AI-generated malicious content

Exploits & CVEs

No major CVE or exploit disclosures reported in the last 24 hours that meet the inclusion criteria.

AI Security

How to encourage smarter AI use in the classroom

Source: MIT Tech Review AI | Risk: MEDIUM | Impacted: K-12 and higher education IT departments, District and campus security teams

Summary: This article is from Making AI Work, MIT Technology Review’s limited-run newsletter examining how to apply LLMs across industries. To receive it in your inbox, sign up here. Chatbots took many schools by surprise upon their release a few years ago. Suddenly, students carried an app in their phones that could magically answer almost any…

Why it matters: Unmanaged student use of AI in school environments can subvert data privacy controls, introduce compliance risks, and create unexpected exposures for institutional IT teams.

Practitioner Perspective

Education IT and security teams have been caught unprepared as generative AI platforms land in the hands of students, who may use them in ways that expose sensitive data or violate usage policy. The lack of clear boundaries has led to haphazard adoption and shadow use, complicating risk management and incident response. Building effective AI classroom policies isn’t about blanket bans, it’s about balancing enablement with visibility and reasonable controls. Engage with educators early to define when and how AI tools can be used safely. Success hinges on translating policy into oversight even when hundreds of students are driving adoption.

Recommended Actions

• Collaborate with school administration to write granular AI usage and monitoring policies for classrooms
• Audit endpoint and cloud application logs for unsanctioned generative AI usage among students

The Download: kids outlearning AI, and space travel agents

Source: MIT Tech Review AI | Risk: MEDIUM | Impacted: Security architects deploying LLM-based controls, Teams responsible for phishing and abuse detection

Summary: This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Kids outlearn AI, and we still don’t know why Teaching a computer to use human language requires an inhuman amount of data. An LLM can easily churn through a hundred thousand times…

Why it matters: Widening gaps between human and LLM learning may mask subtle AI behaviors, making it harder for defenders to anticipate misuse or emerging attack techniques based on human expectations.

Practitioner Perspective

As research underscores differences in how AI models and humans learn, defenders need to resist projecting human-like intuition onto LLM-driven threats or user interactions. This matters because attackers already exploit the non-obvious ways LLMs process and generate content, leading to unpredictable results in prompt injection, social engineering, and phishing campaigns. Update your threat modeling to reflect AI operational quirks and failure modes. If your blue team treats AI outputs like human reasoning, you will miss real exploitation vectors. Experiment with LLMs in a controlled setup to understand how subtle differences can slip past filters that rely on human-like context.

Recommended Actions

• Incorporate AI-native test cases into red team exercises, focusing on unexpected LLM behavioral edge cases
• Adjust user awareness training to include LLM-specific social engineering tactics

Kids outlearn AI, and we still don’t know why

Source: MIT Tech Review AI | Risk: MEDIUM | Impacted: Teams relying on AI tools for policy development, GRC and compliance managers, IT departments automating content moderation

Summary: People have been talking to each other for at least 100,000 years, as best we can tell. And in all that time, there has been only one thing in the world that could learn a human language to perfect fluency: a human child. Now there are two. Four short years after the release of ChatGPT,…

Why it matters: Overreliance on AI-generated content and language models without recognizing their limitations may introduce subtle, hard-to-detect gaps in compliance, safety, or policy implementation.

Practitioner Perspective

Defenders should recognize that LLMs do not learn or generalize in ways that mirror human staff, which affects everything from access policy generation to phishing detection. Assuming AI content mirrors reliable employee output can introduce risk, particularly when controls or reviews are automated or delegated to language models. Adjust your QA and compliance controls to account for the unique error profiles and inferences made by LLMs. Validating output with human review, particularly for high-risk use cases, remains critical. Underestimating these differences erodes both safety and trust in automated controls.

Recommended Actions

• Require human oversight on policy and compliance decisions recommended by LLM tools
• Perform QA sampling on AI-generated guidance or documentation to catch subtle misinterpretations

What We’re Watching

• Monitor for new global SPECTRE-based Linux rootkit campaigns linked to UAT-10147; check telemetry for memory-resident anomalies.
• Immediate policy developments or energy mandate changes in Australia and the EU that could impact hyperscale datacenter operations.
• Uptick in deepfake-for-harassment incidents in schools or public sector organizations; reinforce response playbooks and reporting lines.
• Shadow AI tool API calls from enterprise super-users not visible via approved procurement.
• Any deployment of Ukrainian battlefield-trained AI defense models to civilian or municipal infrastructure operators.

Categories: Artificial Intelligence, Cybersecurity Blog

Tags: AI Security, Data Center Compliance, deepfake, Incident Response, Remediation Debt, Threat Intelligence