Faster Attacks Demand Deeper Resilience
10 articles on faster AI-enabled attacks, agent controls, cyber insurance, supplier risk and trusted recovery.
AI is shortening the time between finding a weakness and trying to exploit it. This week’s stories examine faster attacks, privileged AI agents, exposed APIs and controls that can stop harmful actions quickly. Cyber insurance can help with the financial impact, but its wording and conditions need careful reading. Resilience also depends on the people and systems behind the response. Shared supplier assessments, a risk team drawn from across the business, and security built into daily work can reveal gaps. Tested recovery and verified endpoints matter when a familiar system cannot automatically be trusted. Together, the 10 articles call for faster decisions, limited damage and recovery that restores confidence.

This Week's Articles
- 01CSO
How AI is changing the nature of cyberattacks
AI is accelerating attack speed, scale and experimentation without fundamentally changing attackers’ objectives or attack stages. Examples include GTG-1002’s September 2025 abuse of Claude Code for espionage and JadePuffer’s July 2026 use of an LLM-driven agent to exploit a Langflow vulnerability (CVE-2025-3248) for cloud credentials. Attackers increasingly favour less-restricted models such as GLM and Kimi 3, although model attribution remains uncertain. As exploitation windows shrink, conventional 30-day patching cycles become inadequate, requiring automated triage and agentic defences with human oversight
Why it mattersAutomated triage and human oversight can help teams respond when exploitation windows become shorter than their usual patching cycle.
- 02IT Pro
Duck and coverage: The insurance impact of a cyber attack
Cyber insurance supports financial resilience through incident response, forensics, business interruption, legal support, PR and ransom coverage. Brokers and insurers cite specialist-policy payout rates of 98%–99%, yet fewer than 15% of SMEs with revenue below £5 million are insured. A 2026 Cohesity survey found that only 22% of UK enterprise CEOs believe their policies cover both additional operating costs and lost revenue. Policyholders should scrutinise business-interruption valuation particularly gross-revenue versus net-profit wording and conditions precedent relating to controls such as MFA and patching.
Why it mattersPolicy wording and required security controls affect which costs an organisation can recover after an incident.
- 03CPO Magazine
Every AI Agent Needs a Kill Switch It Can’t Touch
Autonomous AI agents operate like continuously active, privileged employees, often without an independent shutdown mechanism. Prompt injection, behavioural drift or compromised credentials can turn individually authorised actions into unauthorised data exports, cloud deployments or payments. The author estimates that roughly one in five agents in examined environments is untracked. CrowdStrike’s cited 2026 findings: 29-minute average eCrime breakout time, 27 seconds at the fastest underline the need for rapid intervention. Organisations need identity-based controls outside the agent’s own environment to monitor behaviour, revoke credentials and pause actions within seconds.
Why it mattersIndependent monitoring and shutdown controls allow teams to interrupt risky agent actions without depending on the agent to stop itself.
- 04Security Magazine
Security Through Obscurity Is Dead. Long Live the King
Generative AI is compressing the interval between vulnerability disclosure and exploitation from months to hours or minutes, eroding the traditional patching buffer. Cited telemetry shows known actively exploited vulnerabilities rising from approximately 480 to more than 1,740 in a year, against over 380,000 documented flaws. The article advocates an “assume breach” architecture combining zero trust, micro-segmentation, behavioural detection, deception, virtual patching, browser isolation and defensive AI. For boards and CISOs, the investment priority shifts from perimeter complexity towards limiting damage and maintaining recovery readiness
Why it mattersWhen patching cannot keep pace, layered controls can help limit the damage from an intrusion and support recovery.
- 05Forbes
One Risk, Three Lenses: Why Supply Chain Risk Demands That Risk Management, Cybersecurity and IT Work As One
Supply-chain failures are business risks, whether caused by attacks or operational errors. The article cites third-party involvement in 30% of breaches in Verizon’s 2025 DBIR, supply-chain involvement in 15% in IBM’s 2025 report, and the CrowdStrike outage’s impact on 8.5 million Windows devices. Siloed assessments can label vendors “low risk” while overlooking privileged access or automatic updates. Drawing on NIST guidance, it recommends a shared supplier inventory, consistent criticality ratings, contractual security requirements, staged update testing and joint tabletop exercises across risk management, cybersecurity and IT.
Why it mattersA shared supplier view can reveal privileged access and update risks that separate team assessments may miss.
- 06SC World
Racing against the clock: How to stand up to emerging agentic AI threats
Frontier AI is reducing vulnerability discovery and exploit chaining from weeks to hours. Akamai reports a 113% increase in average daily API attacks between 2024 and 2025, API incidents at 87% of surveyed organisations, AI browser-extension adoption among over 40% of enterprise users, and sensitive data in more than 6% of chatbot conversations. Model Context Protocol servers introduce another exposure: hidden instructions can manipulate agents into leaking credentials. Recommended responses include adaptive edge governance, browser controls, brand-authority monitoring and agent autonomy proportionate to how readily actions can be verified and reversed.
Why it mattersAgent permissions and safeguards should reflect how readily their actions can be checked and reversed, alongside API and browser controls.
- 07InformationWeek
Who should be on a CIO's IT risk management team?
IT risk management requires a small, cross-functional council rather than a technical silo. Security, infrastructure and backup teams should work alongside finance, operations and vendor management, with external specialists challenging internal blind spots. The article highlights social engineering as a recurring weakness: organisations often lack an explicitly accountable owner for employee testing and training. A single owner for the risk register should coordinate this work and ensure digital risks are assessed by business impact, not technical severity alone.
Why it mattersA named risk owner and input from across the business can help prioritise digital risks by their actual business impact.
- 08Dark Reading
AI Raises the Stakes for Cyber Resilience
At Commvault’s SHIFT 2026 conference, executives argued that AI-driven attacks and autonomous workflows make continuous resilience a competitive requirement. Recovery must extend beyond restoring systems to restoring clean, trusted data within an isolated environment. AWS emphasised designing AI workflows to “fail safely” through reversible decisions and limited blast radius. Commvault cautioned that AI cannot substitute for foundational asset and identity hygiene. Organisations should also inspect critical suppliers’ actual configurations rather than relying solely on security questionnaires.
Why it mattersClean, isolated recovery and reversible AI actions can limit disruption, while supplier checks need to look beyond questionnaires.
- 09CXO Today
The Invisible Security Paradigm: Embedding Cyber Resilience into Operational Workflows
Drawing on Allcargo Group’s logistics operations, the article explains how interconnected cloud platforms, APIs, devices and suppliers expose shipment visibility and operational continuity, not merely data confidentiality. It advocates risk-led resilience over compliance checklists, prioritising vulnerabilities by exploitability, live exposure and asset criticality rather than severity scores or patch counts. Key measures include asset and identity visibility, zero trust, segmentation, tested containment and recovery, and AI-assisted analysis guided by human judgement. Embedding these practices into everyday workflows makes resilience an operational responsibility rather than a standalone security exercise.
Why it mattersEmbedding visibility, containment and tested recovery into daily operations helps protect essential services.
- 10SecurityInfoWatch
From Known State to Trusted State: A New Approach to Endpoint Security
The article advocates immutable, non-persistent endpoint operating systems to reduce unauthorised changes and prevent their survival across reboots, particularly in VDI, DaaS,SaaS,browser-centric and shared-workstation environments. It cites CrowdStrike’s finding that 82% of detections were malware-free, reinforcing the limitations of malware-focused protection. However, restoring a familiar configuration does not establish trust: the system must be cryptographically verified from firmware upwards through UEFI Secure Boot, Unified Kernel Images and signed, read-only runtime components. The distinction is between returning to a known state and proving that state is trustworthy.
Why it mattersA system returning to its usual configuration does not prove it is safe; verification is needed to establish trust.
