The security stories enterprise IT leaders actually read.
Volume 342, Issue 4 · 20 articles from 6 issues
Only 15 of 154 countries with cybercrime statutes offer any legal protection for good-faith research.
NCC Group's Katharina Sommer mapped global cybercrime law and found only 15 countries have implemented or are considering legal protection for security researchers, out of 154 with cybercrime statutes.
The UK's Computer Misuse Act, enacted in 1990, still makes no distinction between malicious hacking and good-faith research, leaving researchers exposed to fines or prison.
Portugal's 2025 amendment creating a safe haven prompted her to survey further; Argentina, Chile, and Panama have also adopted protections, with Panama covering those who produce hacking tools.
She built the survey by feeding the UN Global Cyberlaw Tracker into an LLM. Her five-point CICIC framework covers conduct, intent, consensus, institution, and conditionality, regulating the activity rather than the actor.
CSA research finds 77% of security stacks now use generative AI, but most teams still block autonomous fixes.
The Cloud Security Alliance's State of AI Cybersecurity 2026 report finds generative AI now appears in 77% of security stacks, alongside high use of agentic AI (67%), supervised ML (67%) and NLP (58%).
Despite this adoption, trust lags: while 89% report visibility into AI reasoning, 74% restrict autonomous action pending better explainability and 86% won't let AI take even small remediation actions without human sign-off.
The report warns that over-reliance on human confirmation risks recreating the alert-backlog problem AI was meant to solve, and finds 85% of security professionals prefer getting SOC capabilities through managed service providers.
Black Hat's recap of USA 2026 identifies the arrival of the agentic age as the event's overarching theme, spanning AI agents, predictive defense, and machine-speed operations.
More than 23,000 members of the cybersecurity community gathered in Las Vegas across six days of Trainings, Briefings, Summits, Keynotes, Arsenal demonstrations, and Business Hall conversations, with attendance up more than 15% year over year.
Discussions of agentic AI that were once theoretical became tangible across research presentations, executive discussions, and community-built demos. Sponsors including ThreatLocker, Microsoft Security, Google Cloud Security, Cisco, SentinelOne, ServiceNow, Wiz, Tenable, and Qualys showed security strategies built around autonomous systems and AI-powered analysis.
Restoring each platform individually does not restore the business; configuration state across platforms is the missing piece.
Aharon Twizer of ControlMonkey argues most enterprises have fragmented resilience: separate vendors for backup, identity, cloud and observability, each able to recover itself but none coordinating a business-level restore.
Resilience is measured by how quickly the business can operate again, and the overlooked dependency is configuration - groups, roles, conditional access policies, application integrations, authentication rules and permissions must return to a trusted state, and the same problem repeats in networking.
Misaligned configuration between interdependent platforms can leave every system technically available while the business still cannot function. He calls for cross-platform configuration visibility and versioned, collectively restorable trusted states.
NIS2 Compliance: Fixing IAM and Access Control Before the 2026 Audit
The NIS2 Directive imposes binding security obligations on EU organizations with penalties up to 10 million euros for non-compliance.
Rather than attempting comprehensive compliance at once, organizations should focus on quick-win controls like credential inventory, centralized vault deployment, and phishing-resistant MFA, implementable in two to four weeks and addressing attack vectors present in 39% of breaches.
Three critical gaps that cause audit failures are unmanaged service accounts and API keys, dormant accounts from poor offboarding, and missing phishing-resistant MFA on privileged access. Auditors require documented evidence through access policies, technical enforcement logs, and exportable audit records.
Passkeys Were Supposed to Be More Secure Than Passwords. Now They're Getting Hacked
Security researchers discovered Pass-ta-key attacks that can extract passkeys from Google Password Manager on Windows, potentially compromising accounts.
However, the vulnerability requires the device to already be compromised by malware and only affects Google's implementation, not passkey technology itself.
FIDO Alliance specifications allow passkeys to be stored locally rather than exclusively in dedicated hardware like TPMs, enabling cloud syncing but creating this specific Windows vulnerability.
Experts note passkeys remain generally secure, and users can mitigate risk by using alternative managers or hardware-based storage like Windows Hello or YubiKeys.
Adding detection tools no longer scales; eliminating unnecessary exposure across digital, physical and human layers does.
The author argues that piling on detection tooling produces operational noise rather than reduced risk, citing that 80% of organizations experienced a cloud security incident in the past year and that systems are probed roughly 2,000 times a week.
Every new cloud service, API and integration expands entry points faster than teams can triage alerts. The proposed reframing is to ask how to eliminate the conditions that allow threats rather than how to detect more of them.
Practically that means continuous asset visibility, decommissioning unused systems, enforcing least privilege across the digital, physical and human access layers, and automating with EASM and CSPM tooling.
Walmart colocates red and blue teams and embeds blue observers in red campaigns to kill win-lose dynamics.
Walmart global VP of security operations Jason O'Dell has restructured purple teaming around trust rather than competition, moving red and blue teams into the same business unit and physical location.
During adversary emulation, two blue team members sit with the red team as trusted agents, checking in real time whether telemetry, logs, and detections fire and ensuring the exercise does not harm the business.
O'Dell's point is that a red team that only attacks and files a report delivers limited value, while one that tells defenders what would have made the attack harder forces both sides to adapt.
The result is incremental improvement rather than a win-lose scoreboard, which also helps retention. The model depends on observers who will not tip off their own colleagues.
Subscription tools starting around $150 a month embed hostile instructions in email, PDFs, and invites.
Proofpoint researchers found threat actors building and selling tools that embed malicious instructions inside emails, documents, calendar invites, and web pages that AI systems process.
These indirect prompt injection attacks manipulate AI agents without requiring any direct user interaction, and subscription offerings start around $150 a month.
Techniques include hidden text in PDFs, concealed prompts in calendar invitations, and instructions embedded in page elements.
The tooling remains largely experimental but signals a shift toward practical exploitation, and the recommended mitigations are restricting agent permissions, monitoring unusual activity, requiring approval for sensitive actions, and rehearsing prompt injection incident response.
Top Cybersecurity Trends to Watch in 2027
Cybersecurity in 2027 will face unprecedented challenges from AI-powered attacks and autonomous agents that access data and make decisions with minimal human oversight.
Organizations must strengthen controls over digital identities for humans, machines, APIs, and AI systems while addressing supply chain vulnerabilities, with third parties involved in 48% of breaches.
Quantum-resistant cryptography migration cannot wait for quantum computers to emerge, as adversaries practice Harvest Now, Decrypt Later attacks.
Defensive strategies must shift from prevention alone toward resilience, combining rapid recovery capabilities with proactive cryptographic preparation and tight control across cloud systems.
Cyber Risk and Incident Response
Cyber incidents now pose significant financial threats to organizations, with research showing 78% of European firms facing up to $500,000 per hour in losses during outages.
Rather than relying solely on prevention, enterprises should adopt proactive resilience strategies including immutable backups, recovery testing, and board-level engagement in incident planning. The evolution of ransomware-as-a-service and flawed encryption tools makes traditional ransom payment approaches increasingly ineffective.
Organizations that invest in recovery preparedness, regularly validate restoration plans, and ensure leadership involvement in tabletop exercises demonstrate superior ability to limit disruption and maintain business continuity when attacks occur.
A Black Hat 2026 vendor survey finds security tools skew toward detection over ones that actually fix problems.
Security researcher Andy Ellis's analysis of roughly 450 vendor booths at Black Hat 2026 finds AI now pervades nearly every product category-Identity, SaaS, AppSec and Data-even when it's not explicit in marketing taglines.
Ellis sorts the market into diagnostic, preventative and corrective tools, finding detection-focused solutions significantly outnumber ones that actually resolve problems.
He suggests this reflects a market dynamic where, as one commenter put it, 'terror is easier to sell than salvation,' since fear-based marketing drives purchases of remediation tools more effectively than straightforward fixes.
Machine identities vastly outnumber human accounts in production, creating security blind spots for defenders.
According to ClearVector's 2026 Identity Intelligence Report, non-human identities including service accounts, execution roles, and vendor credentials comprise 91% of active identities in production cloud environments.
While human activity consists primarily of read-only operations, machine identities execute destructive actions at scale, making compromised credentials especially dangerous.
Only 20% of non-human activity occurs during standard business hours, letting malicious API calls blend into legitimate traffic.
Recent breaches such as Miasma and LiteLLM exploited these weaknesses, and standard authentication and posture tools cannot distinguish legitimate machine behavior from compromised activity.
As AI makes technically sound recommendations cheap, the scarce skill becomes deciding what to do with them.
This op-ed argues that AI now handles much of the analysis and pattern-matching security teams once did, which shifts the weight of the job onto judgment.
A recommendation can be technically correct and still be wrong for the environment: patching a critical flaw immediately makes sense until the affected device is a line controller or a vendor-certified medical device, and blocking a malicious IP makes sense until it is shared CDN infrastructure a business service depends on.
Experienced practitioners carry context no inventory or runbook holds - which unimportant server underpins a critical process, which alarming activity is an authorized red team. The example given is a service account spike that turned out to be quarterly close.
Agents treat llms.txt vendor docs as ground truth and install packages nobody owns - a new supply chain path.
Researchers scanned 6,214 corporate domains belonging to defense contractors and Fortune 500 companies and found 120 references to unregistered packages inside machine-readable documentation files such as llms.txt and llms-full.txt.
After registering some of the unclaimed names, they got callbacks from Fortune 500 environments within hours, showing that AI coding agents had automatically installed and executed the packages.
The quoted diagnosis is that the trust model is broken because agents treat vendor docs as ground truth and the humans supervising them do not question it either. Schneier compares the exposure to SolarWinds-style supply chain risk, noting it differs from prompt injection because the original documentation was written with no malicious intent.
Cisco Talos and CrowdStrike research shows AI is now embedded in daily cybercriminal operations.
Research from Cisco Talos and CrowdStrike documents how threat actors are weaponizing AI to develop malicious code, build fraud infrastructure, and accelerate vulnerability research. Attackers bypass AI guardrails using basic social engineering tactics, and sophisticated groups use AI as a development assistant and system administration tool.
Supply chain attacks increasingly target AI infrastructure, with 87% of identified threats involvingmalicious npm packages.
The window between vulnerability disclosure and active exploitation continues shrinking, with 88% of observed exploitations occurring within 48 hours of a public proof-of-concept release.
August set a record at 398 CVEs resolved, and the binding constraint on patching is now testing and deployment time.
Todd Schell's monthly forecast describes an ongoing patch apocalypse driven by AI-accelerated vulnerability discovery, with August 2026 setting a record of 398 resolved CVEs even though actual exploitation stayed limited to one confirmed exploited flaw and two publicly disclosed ahead of patches.
Notable recent items include chained SharePoint flaws enabling authentication bypass and RCE, an Exchange Server elevation-of-privilege bug, CVSS 10.0 issues in Azure Arc and Exchange Online, and a Microsoft Defender vulnerability with public proof-of-concept code.
His recommendation is to treat network controls as a meaningful defensive layer while risk-driven remediation catches up, and he flags October end-of-support dates for Windows 11 24H2, Server 2012/R2 and Exchange 2016/2019.
Cloud-connected physical security systems are moving onto the CIO's agenda alongside cyber security.
Physical security, once owned by facilities teams, now runs on cloud platforms, connects to enterprise networks and generates operational data, putting it squarely on the CIO's agenda.
Physical and cyber risk have become inseparable: a compromised badge or unsecured entrance can become a network breach, while cyber attacks can disable surveillance and access control.
Legacy on-premise video management systems carry high total cost of ownership through dedicated servers, upgrades and multi-vendor complexity, making cloud-based platforms a logical extension of existing cloud migration programs.
Author Mark Coates argues the data these modern systems produce delivers value well beyond defense.
Noma Labs' 'workflow identity hijacking' lets attackers abuse privileged AI workflows through unauthenticated inputs to steal data.
Researchers at Noma Labs identified 'workflow identity hijacking,' an authorization design flaw in enterprise AI pipelines that decouple the identity of whoever triggers a workflow from the privileges used to execute it.
Attackers send benign-looking requests through unauthenticated entry points such as support inboxes, GitHub issues or web forms, and the workflow executes them with high-privilege service accounts, for example emailing back a finance director's messages.
Experts stress this is an identity issue rather than prompt injection. Recommended defenses include scoped, short-lived delegation tokens, authorization checkpoints treating LLM output as untrusted, separating data retrieval from external responses, and deception assets.
Agentic AI breaks zero trust's per-request model by chaining individually authorized actions into unapproved outcomes.
Zero trust verifies each request, but autonomous agents can walk through several legitimately permitted doors and land somewhere the business never approved.
Practitioners quoted say the framework cannot tell whether an approved agent is still the same agent after new tools, expanded memory or delegated authority are added, so the badge on Friday may not match what was approved Monday.
One estimate holds that roughly 80% of enterprise agents are not on any inventory, and agent-to-agent messages hidden in images create visibility gaps CISOs cannot see into. Suggested fixes include cryptographic delegation modeled on OpenPGP, rate limits, transaction boundaries and reversible decision frameworks.