Coverage: Last 72 hours

Today’s Highlights

Two high-impact security items dominate this cycle: an advanced evasion technique against AI agent skill scanners, and the ‘Bad Epoll’ kernel bug allowing unprivileged-to-root escalation on Linux and Android. Both require immediate focus from defenders managing AI workflows or Linux fleets. The rest of the feed is heavy on emerging AI governance, risk themes, and health sector privacy exposures, but offers limited direct operational guidance.

Table of Contents

1. SkillCloak Lets Malicious AI Agent Skills Evade Static Scanners with Self-Extracting Packing
2. New “Bad Epoll” Linux Kernel Flaw Lets Unprivileged Users Gain Root, Hits Android
3. Doctors’ soaring use of AI scribes prompts Australian government warning over privacy
4. NHS to use AI on its app to direct patients to appropriate services
5. Google DeepMind Unionization Talks Are Off to a Rocky Start
6. ‘It’s smoke and mirrors’: hope turns to fear in Scottish village chosen for AI datacentre
7. What are Britain’s AI growth zones and are the plans feasible or ‘complete bunk’?
8. Revealed: landmark Scottish AI project has no prospect of meeting renewables promise
9. I don’t want a US tech bro as a patron – which is why artists must defend our copyright in the age of AI | Anna Funder
10. AI poses ‘Hiroshima’-style threat to humanity without global rules, says Cooper
11. What’s Kylie’s favourite masking tape? How does Lena Dunham train pigs? It’s all out there – and I’m loving it | Emma Beddington

Top Stories

SkillCloak Lets Malicious AI Agent Skills Evade Static Scanners with Self-Extracting Packing

Source: The Hacker News | Risk: High | Impacted: AI agent orchestration platforms, Organizations deploying third-party AI skills, Security operations teams for AI/ML stacks

Summary: Scanners meant to catch malicious add-on “skills” for AI coding agents can be fooled by a few simple changes that leave the malware working, according to a new study from researchers at the Hong Kong University of Science and Technology. Their strongest trick slipped past every scanner tested more than 90% of the time, and the same team built a runtime checker.

Why it matters: Malicious AI agent skills packaged using advanced self-extracting methods can bypass static security scanners, increasing the risk of undetected code execution within otherwise trusted AI environments.

Practitioner Perspective

Teams managing AI agent skills must recognize that static analysis alone is no longer adequate: threat actors are now actively packaging malicious code in ways that evade even specialized detection tools. This raises the exposure of production AI environments to novel attack chains, where initial skill deployment is treated as trusted by default. Realistically, adversaries will use these bypasses to persist or laterally move, leveraging the trust placed in AI agent infrastructures. You should assume that unsigned or third-party agent skills could be hostile even after scanner review. The priority is to instrument runtime and behavioral controls, not just rely on the hygiene of uploaded artifacts.

Recommended Actions

• Deploy runtime behavior monitoring for all AI coding agent skills, do not rely solely on pre-deployment static scanning
• Block unsigned or unverified skills in your AI skill repository, enforcing provenance checks

New “Bad Epoll” Linux Kernel Flaw Lets Unprivileged Users Gain Root, Hits Android

Source: The Hacker News | Risk: Critical | Impacted: Linux server operators, Android device fleets, Multi-user Linux environments, Container host systems

Summary: A newly disclosed Linux kernel flaw called Bad Epoll (CVE-2026-46242) lets an ordinary user with no special access take full control of a machine as root. It affects Linux desktops, servers, and Android, and a fix is out. Bad Epoll sits in the same small stretch of kernel code where Anthropic’s most powerful AI model, Mythos, recently found a different.

Why it matters: Unprivileged attacker access on Linux or Android systems can be escalated to full root control due to a kernel flaw, potentially enabling lateral movement, privilege escalation, or complete compromise of multi-tenant or exposed workloads.

Practitioner Perspective

Any estate running unpatched Linux kernels or Android deployments faces a direct privilege escalation risk from CVE-2026-46242, known as Bad Epoll. The broad reach covers servers, desktops, and mobile infrastructure, giving attackers a universal local escalation path if they get any foothold. In environments with shared or public access (including containers), this turns minor web app bugs into full system root. You should prioritize patch validation and incident response readiness for post-exploitation activity linked to this CVE. Assume exploit availability if your stack remains unpatched.

Recommended Actions

• Deploy kernel updates addressing CVE-2026-46242 (Bad Epoll) across all Linux systems, including production servers and VMs
• Push vendor-provided Android security updates for CVE-2026-46242 to managed mobile fleets

Emerging Signals

Google DeepMind Unionization Talks Are Off to a Rocky Start

Source: The Verge AI | Risk: Medium | Impacted: AI industry labor relations stakeholders, DeepMind workforce, Organizational risk managers

Summary: During negotiations on Wednesday, employees voiced frustrations with what they consider an unwillingness among executives to engage meaningfully with the prospect of unionization.

Why it matters: Difficult union negotiations at prominent AI labs can signal upcoming internal disruptions and management changes that may impact AI research continuity and operational security posture.

Practitioner Perspective

Labor instability during transformation phases in technologically critical companies often leads to operational friction, slower vulnerability response, or loss of knowledge regarding internal controls. Security teams should build contingency for service availability and knowledge handoff if staff attrition or work disruptions worsen. Monitor for organizational signals indicating upcoming leadership or process shifts that could temporarily weaken governance maturity.

Recommended Actions

• Monitor for organizational changes affecting security operations within critical AI vendors such as DeepMind
• Strengthen third-party risk assessments related to supplier stability

Exploits & CVEs

No additional exploits or CVEs reported in this cycle beyond Top Stories.

AI Security

‘It’s smoke and mirrors’: hope turns to fear in Scottish village chosen for AI datacentre

Source: The Guardian | Risk: Medium | Impacted: Local communities near AI infrastructure projects, Policy analysts, Supply chain assessors

Summary: Suspicions grow in Lanarkshire that local people have been misled on supposed benefits of the huge development. Revealed: landmark Scottish AI project has no prospect of meeting renewables promise. What are Britain’s AI growth zones and are the plans feasible or ‘complete bunk’? The promise was that a Scottish community would be transformed by massive investment and empowered to chase.

Why it matters: Local communities could face direct economic or environmental risks from overpromised AI infrastructure projects that underdeliver on jobs or sustainability.

Practitioner Perspective

Wherever AI datacentres are deployed, local engagement and due diligence are essential. Vague or misleading promises regarding sustainability or economic uplift can quickly turn communities against critical technology infrastructure. Project leaders and public sector partners must provide transparency and set realistic expectations to minimize risk of backlash or regulatory delay.

Recommended Actions

• Assess risk frameworks for local IT project community outreach and accuracy in marketing claims
• Coordinate with local public authorities regarding impact reports and project dependencies

What are Britain’s AI growth zones and are the plans feasible or ‘complete bunk’?

Source: The Guardian | Risk: Medium | Impacted: UK technology sector, Public sector decision makers, AI infrastructure planners

Summary: Lanarkshire datacentre run by renewables and creating thousands of jobs not achievable by 2030, Guardian investigation finds. Revealed: landmark Scottish AI project has no prospect of meeting renewables promise. ‘It’s smoke and mirrors’: hope turns to fear in Scottish village chosen for AI datacentre. The Guardian has examined government plans to build Britain’s AI infrastructure for the future, finding some.

Why it matters: Overly optimistic projections on AI infrastructure scale-out may undermine trust and proper planning across the wider digital economy if targets are not met.

Practitioner Perspective

Long-term growth in the AI sector depends on transparent, achievable infrastructure planning. Security and operational teams should expect pressure to accelerate or shoehorn projects onto unproven or politically motivated timelines. Insist on honest resource appraisals and business continuity pressures before scaling critical AI infrastructure.

Recommended Actions

• Prepare contingency plans for project delays or environmental compliance shortfalls
• Request up-to-date project plans and third-party feasibility studies before major investments

Revealed: landmark Scottish AI project has no prospect of meeting renewables promise

Source: The Guardian | Risk: Medium | Impacted: Energy planners, Datacentre operators, Environmental compliance officers

Summary: Exclusive: Government and developers privately acknowledged Lanarkshire datacentre site had power provision ‘issue’. ‘It’s smoke and mirrors’: hope turns to fear in Scottish village chosen for AI datacentre. What are Britain’s AI growth zones and are the plans feasible or ‘complete bunk’? A landmark AI development billed as delivering jobs and prosperity has misrepresented its plans to channel a nuclear.

Why it matters: Inadequate power and sustainability planning for large AI datacentres may create new environmental and grid stress risk.

Practitioner Perspective

Failing to meet renewables or power delivery promises opens not only reputational risk but also hard limitations on future AI workloads. Security and facilities planners should not assume promised power sources or sustainability metrics are realistic until proven. Create fallback plans for grid, regulatory, and reputational disruptions.

Recommended Actions

• Incorporate hard audit points on energy sourcing and grid impact for AI buildouts
• Escalate critical questions on environmental compliance to executive risk steering

I don’t want a US tech bro as a patron – which is why artists must defend our copyright in the age of AI | Anna Funder

Source: The Guardian | Risk: Medium | Impacted: Creative professionals, Copyright officers, Policy advocates

Summary: US companies hoping to make fortunes from AI want the creative product of our country to be available to them for free, or for peanuts. Words fail me. Last week I went to Canberra with a group of other people who live by selling our creative work. We make the books, songs, artwork, First Nations artworks, films, music and TV.

Why it matters: Rapid AI content deployment is intensifying creative community pushback, potentially leading to stricter copyright protections or legal challenges against large AI vendors.

Practitioner Perspective

Concerns from creators about AI’s use of original works will likely cascade into new policy measures and enforcement regimes. AI organizations must stay ahead of emerging copyright law, ensure dataset provenance, and communicate standards for material usage.

Recommended Actions

• Validate datasets for legal clarity before training or deployment
• Monitor legislative and policy shifts regarding copyright compliance

AI poses ‘Hiroshima’-style threat to humanity without global rules, says Cooper

Source: The Guardian | Risk: Medium | Impacted: International policy makers, AI risk analysts, Global security agencies

Summary: Exclusive: Foreign secretary warns of combined risks of AI, climate crisis, irregular migration and foreign interference. Artificial intelligence poses a “Hiroshima”-style risk to humanity if governments do not agree to curb how it is developed, the foreign secretary has warned. Yvette Cooper urged countries, including the US and China, to agree international rules for AI, telling the Guardian she believes.

Why it matters: Calls for global regulation could accelerate multilateral AI risk management policies, directly shaping system development and attack surface priorities for all major stakeholders.

Practitioner Perspective

Security, risk, and compliance teams must prepare for new international requirements and potential controls on both offensive and defensive uses of AI. Advance preparation on audit, reporting, and cross-border standards will reduce the pain of sudden regulatory change.

Recommended Actions

• Map out current uses of AI that intersect with global risk categories
• Track international policy movements and upcoming compliance mandates

What’s Kylie’s favourite masking tape? How does Lena Dunham train pigs? It’s all out there – and I’m loving it | Emma Beddington

Source: The Guardian | Risk: Low | Impacted: Content moderators, Digital culture analysts, Social media teams

Summary: The more I learn about celebrities and their odd passions, the more encouraged I am. So much for AI drowning us in a flood of bland ‘tasteslop’. The internet, as we know, is now a depressing hellhole where everything is a terrifying shot of cortisol straight into the eyeballs or AI slop, interspersed with adverts for protein. So may I.

Why it matters: Cultural resilience to AI-driven content could shape demand and trust in both human-created and AI-generated media, impacting digital experience strategies.

Practitioner Perspective

Security and content governance professionals should remain aware of shifting public sentiment regarding AI content, as these perceptions can affect user trust and policy debate. Balance between moderation and creative engagement is required going forward.

Recommended Actions

• Track trends on AI fatigue and emerging content preferences
• Adjust content moderation guidelines as digital culture adapts

Doctors’ soaring use of AI scribes prompts Australian government warning over privacy

Source: The Guardian | Risk: High | Impacted: General practice clinics using AI scribes, Healthcare IT administrators, Health sector privacy officers

Summary: Exclusive: With the technology fast becoming popular in GP surgeries, regulators are monitoring its implementation and potential pitfalls. The federal health department has raised concerns about the use of AI scribes by doctors as the health regulator considers the need.

Why it matters: The unchecked adoption of AI scribes in healthcare creates exposure to new data privacy risks, especially if sensitive patient information is processed or stored by third-party AI services outside established regulatory compliance frameworks.

Practitioner Perspective

Healthcare providers implementing AI scribe technologies are now squarely on regulatory radar for private data handling practices. The pace of AI adoption in clinical workflow increases the sensitivity of operational risk, because patient records may be processed by AI vendors with insufficient privacy controls or unknown data retention policies. Security teams cannot treat these integrations as benign: you must establish clear vendor data pathways, audit for shadow IT AI tools, and treat all unvetted scribe APIs as likely to leak or mishandle protected health information. The moment to tighten contractual security and monitoring requirements is now, not post-incident.

Recommended Actions

• Review and update contractual agreements with AI scribe vendors to enforce compliance with healthcare data privacy regulations
• Conduct technical validation of data flow for deployed AI scribe integrations, ensure patient data does not leave approved jurisdictions

NHS to use AI on its app to direct patients to appropriate services

Source: The Guardian | Risk: Medium | Impacted: NHS app product teams, Health IT security engineers, Patient data privacy teams

Summary: Update in England expected to reach about 200,000 patients over the next year as part of £10bn package to overhaul NHS systems. The NHS will begin using AI on its app to direct patients to the appropriate services, it has been announced. The tool will be used to triage patients and to ascertain if they should be allocated a GP.

Why it matters: Automated AI triage in health applications can introduce misrouting or disclosure risks if algorithms are not thoroughly tested or if data handling does not meet sector-specific security baselines.

Practitioner Perspective

As health orgs move patient triage to AI-powered app features, day-to-day risk increases for leakage of sensitive clinical data and accidental triage misclassification, directly affecting outcomes and regulatory standing. Defenders should scrutinize any changes to backend workflow, especially where new AI-driven decision logic is integrated into legacy EHR or appointment systems. You cannot assume the underlying AI model is robust or that its cloud endpoints inherit the same security posture as on-prem clinical infrastructure. Ensure robust audit and fallback procedures are defined before rollout.

Recommended Actions

• Conduct security review of NHS app AI triage code and associated data flow changes prior to production rollout
• Audit access control and data encryption for backend servers hosting triage logic and patient information

Defensive Actions

• Deploy runtime behavior monitoring for all AI coding agent skills, do not rely solely on pre-deployment static scanning
• Block unsigned or unverified skills in your AI skill repository, enforcing provenance checks
• Adopt threat hunting focused on suspicious skill unpacking or self-extracting behaviors in AI agent environments
• Update onboarding and review workflows for AI skills to mandate dynamic sandboxing before production approval
• Deploy kernel updates addressing CVE-2026-46242 (Bad Epoll) across all Linux systems, including production servers and VMs
• Push vendor-provided Android security updates for CVE-2026-46242 to managed mobile fleets
• Audit privileged activity logs on Linux hosts for anomalous escalation attempts post-CVE disclosure window
• Review and update contractual agreements with AI scribe vendors to enforce compliance with healthcare data privacy regulations
• Conduct technical validation of data flow for deployed AI scribe integrations, ensuring patient data does not leave approved jurisdictions
• Conduct security review of NHS app AI triage code and associated data flow changes prior to production rollout

What We’re Watching

• Ongoing developments in runtime evasion tactics against AI agent scanners
• Rapid patch uptake and post-exploitation monitoring for the Bad Epoll kernel vulnerability
• Regulatory movement and privacy controls for AI healthcare tools
• Policy deliberations shaping the future of large-scale AI infrastructure projects
• Shifts in public sentiment, labor stability, and digital cultural engagement in response to accelerated AI integration

Categories: Artificial Intelligence, Cybersecurity Blog

Tags: AI Security, Healthcare Privacy, Linux, Threat Intelligence, Vulnerabilities