What Is a Zero-Day Attack? Risks, Examples & Protection Solutions (2027)

Somewhere right now, a flaw exists in software your business depends on, and the people who built it do not know about it. Sometimes an attacker finds it first. When that happens there is no patch to install, no signature to download and no advisory to read, because defenders have had zero days to prepare. That gap is what makes a zero-day attack so unsettling. It is also why the response that works is rarely a single product. It is a layered habit of reducing exposure, watching closely, and being ready to act the moment a mitigation or fix appears.
TL;DR
Definition: A zero-day attack exploits a vulnerability that the vendor had not fixed, or did not know about, when the attack happened. The flaw is the zero-day vulnerability; the code that abuses it is the zero-day exploit.
Scale: Google Threat Intelligence Group counted 90 zero-days exploited in the wild in 2025, up from 78 in 2024. That counts vulnerabilities, not victims, and only cases that were detected and disclosed.
Speed beats prediction: You cannot patch an unknown flaw, so readiness comes from fewer exposed systems, strong visibility, fast mitigation and practiced response.
No single control wins: EDR, WAF, SIEM, segmentation, backups and authorized testing each cover different moments. None guarantees prevention or detection.
Suspected exploitation is an incident: Patching closes the door but does not evict someone already inside. Investigate first, then verify.
What is a zero-day attack? (Quick answer)
A zero-day attack is a cyberattack that exploits a software, firmware or hardware vulnerability before the vendor has released a fix, or before defenders knew it existed. Attackers abuse the flaw with a zero-day exploit while victims have no patch to apply, which makes early detection, mitigation, and containment critical.
![]() Could Your Web Application or API Have Overlooked Security Weaknesses? Zero-days are unknown by definition, but many real intrusions still start with weaknesses that can be found and fixed in advance: broken access control, exposed API endpoints, risky configurations and weak session handling. Articsledge arranges scoped penetration testing for web applications, APIs and SaaS platforms, delivered under agreed scope and written authorization. It strengthens your defenses; it cannot promise to find every flaw. |
Table of Contents
What Is a Zero-Day Attack?
A zero-day attack happens when someone exploits a hardware, firmware or software vulnerability that the vendor has not yet fixed. NIST's glossary describes it as an attack that takes advantage of a previously unknown weakness. Every internet-facing service adds to your attack surface, and any of them can carry such a weakness. In everyday security language, three related terms are worth separating:
Zero-day vulnerability: the weakness itself, such as a memory-corruption bug or an authentication flaw that has no fix available.
Zero-day exploit: the technique or code that turns that weakness into access, data or control.
Zero-day attack: the actual use of that exploit against a target.
The name comes from the idea that defenders have had zero days to fix the problem. Usage varies, though. Some writers use the term for flaws unknown to the vendor, others for flaws exploited before a fix existed, and some loosely for any recently disclosed flaw. Google Threat Intelligence Group counts a zero-day as a vulnerability maliciously exploited in the wild before a public patch was available. When you read a report, check which meaning it uses and when the flaw became public.
A simple example
Imagine a company runs a file-transfer server exposed to the internet. A criminal group discovers that a crafted request lets anyone read stored files without logging in. The vendor has not heard about it. The attackers quietly copy data for days. Until the vendor learns of the problem and ships a workaround or patch, the normal patch cycle cannot protect that company. Once the fix is published, the same flaw becomes an n-day: still dangerous, but now defenders have something to act on. This scenario is illustrative; the MOVEit Transfer case later in this guide shows a documented one.
How Does a Zero-Day Attack Work?
Zero-day attacks follow a lifecycle that looks different depending on whether you stand with the attacker, the vendor or the defender. The model below is simplified, and the real gaps between stages range from hours to years.

Weakness exists. A coding mistake, design flaw or insecure default sits unnoticed. Most software carries latent bugs, and only a fraction are ever weaponized.
Attacker finds it. A researcher, criminal group, state-linked team or commercial spyware vendor discovers the flaw and builds a reliable exploit, often by chaining several bugs.
Exploitation begins. The attacker uses the exploit in targeted or mass campaigns. No fix exists, so ordinary patching offers no protection.
Defenders notice. A victim's logs, a security vendor's telemetry, a crash report or an incident responder's investigation reveals the activity.
Disclosure. The vendor confirms the flaw and usually assigns a CVE (Common Vulnerabilities and Exposures) identifier. Timing varies, and some vendors disclose only when a patch is ready.
Mitigation, then patch. Vendors often publish workarounds first, such as disabling a feature or applying a configuration change, followed by a fix.
Remediate and verify. Defenders patch, then check whether the system was already compromised, rotate credentials, and confirm the fix worked.
Two points deserve emphasis. First, a vulnerability can remain a zero-day for months or years before stage 4. Second, stage 7 is where many organizations stop too early. Installing a patch ends the exposure but does not remove an attacker who got in during stages 3 to 6. A vulnerability assessment and retesting can help verify that a fix worked, but they are not a substitute for investigating whether intrusion already occurred.
Zero-Day vs. N-Day vs. Zero-Click
Security teams use several overlapping labels, and mixing them up leads to bad decisions. The table separates what each term actually describes.
Term | What it describes | Fix available? | Publicly known? | User interaction needed? |
Zero-day | A flaw exploited before a fix exists, or before the vendor knew of it | No, at the time of exploitation | Not at first | Depends on the flaw |
N-day | A flaw with a public fix or disclosure, exploited afterwards | Yes | Yes | Depends on the flaw |
Zero-click | An exploit that needs no action from the victim | Either | Either | None |
Unpatched, known exploited | A flaw with confirmed attacks that the victim has not yet fixed | Often yes | Yes | Depends on the flaw |
Unknown, not yet exploited | An undiscovered or unreported flaw with no known abuse | No | No | Depends on the flaw |
These categories describe different attributes, so they overlap. A zero-click exploit can be a zero-day, as in the iMessage case covered later, or an n-day. A zero-day may require a click on a malicious link. A CVE identifier does not prove that a flaw was exploited, and an entry in the CISA Known Exploited Vulnerabilities catalog does not prove it was a zero-day, since many entries were added long after patches existed.
In practice, n-days are where patch discipline pays off, and zero-days are where mitigation, detection and containment matter. Speed matters for both. In one case documented by GTIG, exploitation of BeyondTrust's CVE-2026-1731 began within four days of public disclosure. The window between a fix and your deployment of it is a risk you control.
How Zero-Day Exploitation Unfolds
Understanding how exploitation unfolds helps defenders place controls at each step. The outline below is deliberately high-level, mapped to MITRE ATT&CK where useful, and omits operational detail.
Initial access. Attackers favor internet-facing systems; ATT&CK describes this as Exploit Public-Facing Application (T1190). Client-side routes reach users through web pages, documents or messages, often paired with social engineering or phishing.
Exploit chaining. A single bug rarely gives full control of a hardened modern platform. Skilled operators chain several, for example one flaw to run code inside a restricted process and another to break out of it.
Privilege escalation. Once code runs, a second weakness or a misconfiguration may raise the attacker from an ordinary account to administrative or system rights.
Persistence. To survive reboots and patches, intruders may create accounts, schedule tasks, change startup items or leave a web shell behind. Stealthy implants such as a rootkit or fileless malware make this harder to spot.
Lateral movement and goals. With stolen credentials and trusted network paths, attackers move to higher-value systems to steal data, spy or deploy ransomware. Some campaigns, such as mass data theft from file-transfer software, never leave the first server.
The defensive lesson is that each step is a chance to interrupt the chain. Fewer exposed services, least privilege, segmentation and good logging do not need to know which bug was used. They limit what any bug can achieve.
Common Targets and Attack Surfaces
Zero-days follow value. Attackers invest in flaws in technology that is widely deployed, internet-facing or trusted with sensitive access. The Google Threat Intelligence Group 2025 review gives a useful snapshot of where exploitation was observed. Its categories overlap, so the figures below do not add up to 90:
Operating systems. 39 of the 90 exploited zero-days (44%), the largest category in the 2025 data.
Enterprise technology. 43 zero-days (48%), a record share, including 21 in security and networking products, 14 of them on edge devices. Gateways, VPNs and firewalls sit on the network boundary and often have limited endpoint monitoring.
Mobile. 15 zero-days, up from 9 in 2024, including messaging routes that need no user action.
Browsers. Under 10% of the total, a historic low according to GTIG, helped by sandboxing and fast automatic updates. They remain targets for espionage.
Three surfaces matter beyond what those counts capture:
Web applications and APIs. Custom code seldom receives a CVE, yet flaws such as broken access control or business-logic errors can behave like zero-days for the owner: unknown, exploitable and unfixed. The OWASP Web Security Testing Guide is a standard reference for testing them.
Cloud and identity. Identity providers, management consoles and access tokens give broad reach, and cloud security misconfigurations multiply the impact of any single flaw.
Dependencies and third parties. A flaw in a library, a managed file-transfer product or a supplier's platform becomes your exposure. Keep an inventory that includes open-source components.
Business and Individual Risks of Zero-Day Attacks
Impact depends on what a flaw exposes, not on how new it is. The most common outcomes are:
Data theft and extortion. Mass-exploitation campaigns against file-transfer or business software can copy data from many organizations at once, as the MOVEit case below shows.
Business interruption. Emergency mitigations, forced shutdowns and rebuilds can take systems offline. In early 2024, CISA directed US federal agencies to disconnect affected Ivanti products until they could be rebuilt and upgraded.
Account and identity compromise. Edge and identity flaws often expose credentials and sessions, which can outlive the patch unless they are reset.
Ransomware opportunities. A zero-day can provide the first foothold that a ransomware crew later monetizes, which is why containment speed matters.
Response and recovery costs. Investigation, rebuilds, notification and legal work add up. Costs vary too widely for a universal figure to be honest, so this guide does not quote one.
Legal and regulatory duties. Breach-notification and security-measure obligations differ by jurisdiction and sector. In the EU, NIS2 raises expectations for covered entities, and cloud compliance frameworks add their own. Ask qualified counsel which duties apply to you.
For individuals, the realistic risks are account takeover, device compromise and spyware on phones and browsers. Update the operating system, browser and apps promptly, restart devices after updates, remove software you no longer use, and use unique passwords with phishing-resistant sign-in where offered. People at elevated risk, such as journalists or civil-society workers, can consider extra hardening such as Apple's Lockdown Mode.
![]() Turn Application Exposure Into Scoped, Prioritized Findings Knowing that exposure exists is not the same as knowing where it is. A scoped assessment of your web application, API or SaaS platform can turn a vague worry into specific, prioritized findings your team can act on. Articsledge arranges engagements after agreeing scope and written authorization. Remediation work and retesting are separate scopes, and results reflect only what was in scope and tested. |
Documented Real-World Zero-Day Examples
Not every famous breach is a zero-day. Many mass compromises involve n-days that victims had not yet patched, so each case below is labeled by what was publicly established about the flaw at the time of exploitation. Chronologies come from the cited primary sources; where a source did not state a date, this guide leaves it out.
Stuxnet (2010)
The Symantec W32.Stuxnet Dossier describes malware that exploited four unpatched Microsoft vulnerabilities and fingerprinted a specific industrial control system. It showed that specialized, isolated environments can be reached through removable media and internal propagation. Lesson: control removable media and segment operational technology.
Microsoft Exchange, ProxyLogon (2021)
Microsoft reported on March 2, 2021 that four vulnerabilities (CVE-2021-26855, CVE-2021-26857, CVE-2021-26858 and CVE-2021-27065) in on-premises Exchange Server had been used in limited, targeted attacks before fixes were released, attributing the activity to a group it calls HAFNIUM. Updates shipped the same day, with a mitigation tool and investigation guidance following. Lesson: after patching, check whether intruders were already present.
Apple iMessage, BLASTPASS (2023)
Citizen Lab found a zero-click chain, CVE-2023-41064 and CVE-2023-41061, on an iPhone running iOS 16.6 belonging to an employee of a civil-society organization. Malicious images in PassKit attachments required no action from the victim, and the research was linked to NSO Group's Pegasus spyware. Apple released fixes around September 7, 2023. Lesson: high-risk users benefit from fast updates and hardened modes.
MOVEit Transfer (2023)
CISA's advisory AA23-158A describes CVE-2023-34362, an SQL injection zero-day in Progress Software's MOVEit Transfer web application. Open-source reporting cited by CISA places exploitation from May 27, 2023, and attackers used a web shell to steal data from MOVEit databases. The advisory does not describe encryption in this campaign. Lesson: file-transfer and supplier software can expose your data even when your own systems are sound.
Ivanti Connect Secure (2023 to 2024)
Mandiant reported zero-day exploitation of CVE-2023-46805 and CVE-2024-21887 as early as December 2023, with Ivanti disclosing and publishing mitigations on January 10, 2024 while patches were still in development. CISA's emergency directive later required disconnection and rebuilds, and warned defenders to assume sophisticated actors may achieve persistence that survives a factory reset. Lesson: edge devices need visibility, mitigation plans and rebuild procedures.
Google Chrome, CVE-2025-2783 (2025)
Kaspersky detected the campaign it calls Operation ForumTroll in mid-March 2025. The Chrome sandbox-escape flaw was chained with a second exploit that Kaspersky did not obtain, and the fix shipped on March 25, 2025 in Chrome 134.0.6998.177 and .178. Patching the sandbox escape disrupted the chain. Lesson: update browsers promptly and restart them.
Oracle E-Business Suite, CVE-2025-61882 (2025)
Oracle's October 4, 2025 alert describes a remote-code-execution flaw reachable over HTTP without credentials, and lists indicators of compromise for observed activity. GTIG's review counts a campaign by a financially motivated group associated with Cl0p against Oracle E-Business Suite among 2025 zero-days. Lesson: business suites exposed to the internet deserve the same scrutiny as edge devices.
Zero-day incident comparison
Incident | Product and CVE | Zero-day when exploited? | Key chronology | Observed impact |
Stuxnet (2010) | Windows, several flaws | Yes, four unpatched flaws per Symantec | Analyzed publicly in 2010 and 2011 | Targeted industrial control system |
ProxyLogon (2021) | On-premises Exchange, four CVEs | Yes, before fixes | Patches and advisory March 2, 2021 | Limited, targeted attacks first, per Microsoft |
BLASTPASS (2023) | iOS and iMessage, CVE-2023-41064 and CVE-2023-41061 | Yes, zero-click | Found on a device, fixed around September 7, 2023 | Spyware on a targeted individual's phone |
MOVEit Transfer (2023) | CVE-2023-34362 | Yes, per CISA | Exploitation from May 27, 2023 per open-source reporting | Data theft through a web shell |
Ivanti Connect Secure (2023 to 2024) | CVE-2023-46805 and CVE-2024-21887 | Yes, per Mandiant | Exploitation from December 2023, mitigations January 10, 2024, patches later | Credential capture and web shells |
Chrome (2025) | CVE-2025-2783 | Yes, per Kaspersky | Mid-March 2025 detection, fix March 25, 2025 | Targeted espionage chain |
Oracle E-Business Suite (2025) | CVE-2025-61882 | Counted as a zero-day by GTIG | Oracle alert October 4, 2025 | Extortion-linked campaign per GTIG |
Source notes and limits. Dates reflect what each source states publicly, and exploitation may have begun earlier than reported. Impact descriptions are limited to what the cited organizations observed. Last verified: November 23, 2026.
What Current Research Shows
Counting zero-days is harder than it sounds, so every figure needs its definition, window and caveats. GTIG's 2025 zero-day review, published March 5, 2026, counted 90 zero-days exploited in the wild in 2025, against 78 in 2024 and a record 100 in 2023. It defines a zero-day as a vulnerability maliciously exploited before a public patch was available, covers data to December 31, 2025, and notes that counts can change as past incidents are discovered. For the first time, GTIG attributed more zero-days to commercial surveillance vendors than to traditional state-sponsored espionage groups. It also found at least 10 zero-days linked to China-nexus espionage groups and 9 to financially motivated actors, and all 2025 zero-days in its dataset now have patches.
GTIG's later analysis, Vulnerability Discovery and Exploitation Trends in the AI Era (September 30, 2026), covers January 2025 to August 2026. It reports roughly doubled monthly CVE disclosures, exploited vulnerabilities rising from an average of 10.5 per month in 2025 to 18 per month in early 2026, and zero-days from about 8 to about 11 per month. Only 0.23% of 2026 disclosures, about 1 in 431, were observed exploited.
The table keeps the measures separate, since they are often confused.
Measure | What it counts | Latest figure cited | Window | Caution |
Exploited zero-days | Vulnerabilities exploited before a public patch | 90 | Calendar year 2025 | Detected and disclosed cases only |
Exploited vulnerabilities, all | Zero-days plus n-days with observed exploitation | 127 in 2025, 141 in January to August 2026 | Different lengths of period | Not directly comparable without annualizing |
CVE disclosures | Published vulnerability records | About 10,740 in August 2026 | Monthly | Automatic assignment inflated counts |
Exploitation share | Disclosed CVEs seen exploited | 0.23% of 2026 disclosures | January to August 2026 | Small base, sensitive to single campaigns |
Read these numbers with three limits in mind. Reporting bias means well-monitored products and well-resourced vendors appear more often. Counts measure vulnerabilities, not attacks or victims, and one flaw can drive a mass campaign. Finally, GTIG itself describes its AI-discovery attribution as heuristic and the data as early indicators, so treat trend language as informed estimates rather than settled fact.
Detecting Possible Zero-Day Activity
Because no signature exists for a flaw nobody has disclosed, detection of zero-day activity usually relies on behavior and context rather than known indicators. No tool catches everything, so the aim is to make intrusions noisy, short and well recorded.
Endpoint behavior (EDR and XDR). Endpoint protection can flag unusual process chains, such as a web server process launching a command shell or a document reader spawning scripts. XDR correlates endpoint, identity, email and network signals. Neither works where an agent cannot run, which includes many appliances.
Logs worth enabling. Authentication events, administrator changes, new accounts and tokens, web server and API logs, process creation, outbound connections and DNS queries. Keep them long enough: zero-days are often recognized weeks after exploitation began.
SIEM correlation. A SIEM joins signals from many sources, for example a successful login from a new country followed by a new administrator account and a large outbound transfer.
Application and API events. Watch for unexpected file writes in web directories, new files in upload folders, bursts of errors, and access-control failures that suddenly succeed.
Edge-device visibility gaps. Forward appliance logs to a central store, use vendor integrity-check tools, and alert on unexpected administrator sessions or outbound connections from the device itself.
Network telemetry. Unusual destinations, regular beaconing patterns and large transfers can reveal activity that endpoint tools never see.
Indicators and hunting. Indicators of compromise age quickly and mostly describe past incidents. Structured hunting for behaviors, supported by threat detection and response practices and AI threat detection where it has tuned baselines, finds what rules miss.
Triage and false positives. Define in advance what counts as suspected versus confirmed, who decides, and how fast. An alert queue that nobody reads is not a control.
A quiet dashboard does not prove safety. Treat detection as probabilistic and pair it with the preventive controls below.
Prevention and Risk Reduction
Defense in depth means no single failure decides the outcome. These measures work whether or not you know which vulnerability will be used:
Know what you run. Maintain an inventory of internet-facing assets, versions, owners and third-party dependencies. You cannot mitigate a flaw in a product you did not know you had.
Shrink exposure. Remove unused services, place management interfaces behind controlled access, and review your security infrastructure for anything reachable from the internet that need not be.
Apply least privilege. Limit service accounts, administrator rights and API tokens so a compromised component cannot reach everything.
Use phishing-resistant MFA where appropriate. Prioritize administrators and remote access, and support users with phishing simulation and clear reporting channels.
Segment the network. A compromised web server should not have a direct path to databases or identity systems.
Harden configurations. Follow vendor hardening guides, disable unused features and change defaults.
Patch with a triage method. CVSS describes severity. EPSS estimates the probability that a published CVE will see exploitation in the next 30 days and says nothing about undisclosed flaws. CISA's KEV catalog lists vulnerabilities with known exploitation. Combine them with asset exposure, and set an emergency timeline for known exploited flaws on internet-facing systems.
Prepare mitigations before you need them. Know how to disable a feature, restrict access or isolate a system quickly while waiting for a fix.
Use WAFs and virtual patching with realistic expectations. Rules can block known exploit patterns at the HTTP layer and buy time, but they cover only certain traffic, can be bypassed, and are not a substitute for fixing the flaw.
Harden applications and the build process. Validate input, review code, scan dependencies and track components. Mature teams can add a bug bounty program to widen coverage.
Keep recoverable backups. Use offline or immutable copies, test restores regularly, and consider backup as a service where internal capacity is limited.
Rehearse. Run tabletop exercises for a scenario such as an edge-device zero-day, and schedule authorized testing of critical applications.
![]() Complement Preventive Controls With Authorized Application and API Testing Hardening, segmentation and patching reduce exposure, but they rarely prove how your own application behaves under attack. Authorized testing of web applications and APIs can reveal access-control gaps, risky configurations and business-logic weaknesses that preventive controls missed. Articsledge arranges scoped engagements with written authorization. Testing complements patching, monitoring and response and does not replace them. |
What to Do When Exploitation Is Suspected
A vulnerable product is not the same as a compromised one. Separate exposure (an affected version is present), suspected exploitation (signs of attempts or odd behavior) and confirmed compromise (evidence of attacker access). NIST SP 800-61 Rev. 3, which aligns incident response with the NIST Cybersecurity Framework 2.0, treats response as part of ongoing cyber risk management rather than a one-time event. A practical sequence looks like this:
Triage and declare. Classify the situation, name an incident lead and open the escalation path to management, legal and communications.
Preserve evidence. Capture logs, relevant memory and disk images or vendor-recommended snapshots before rebuilding. Record clock settings so timelines line up.
Contain, knowing the trade-offs. Isolate affected systems, block malicious addresses and disable the exposed feature. Weigh evidence preservation against downtime, and do not wipe a system before capturing what you need.
Inventory what is affected. Find every instance and version, and map what each could reach.
Follow vendor and government advisories. Apply mitigations and patches as directed, and watch for superseding updates. In the Ivanti case, mitigations came first, patches followed, and later updates replaced them.
Reset credentials and sessions. Rotate passwords, keys, certificates and tokens, and revoke active sessions. CISA's Ivanti directive required reissuing certificates, keys and passwords before affected products returned to service.
Hunt for persistence. Look for new accounts, scheduled tasks, web shells and unexpected binaries. For some edge devices, a full rebuild is safer than cleaning.
Recover and verify. Restore from known-good sources, retest the fix, and monitor closely for recurrence.
Meet legal and contractual duties. Involve counsel early, because notification clocks may already be running.
Learn. Hold an after-action review and update inventories, logging, playbooks and exercises.
If you suspect active compromise, engage your incident response team or a specialist incident-response provider immediately. Penetration testing services are not emergency response.
Protection Solutions Compared
No tool covers every moment of a zero-day. The matrix below is a qualitative synthesis of how each control category typically contributes before a flaw is disclosed, while it is being exploited, and after a patch exists. It is not a ranking and carries no pricing, because outcomes depend on deployment, tuning and staffing.
Control | Purpose | Before disclosure | During exploitation | After a patch | Limits and best fit |
EDR | Monitors endpoint behavior and supports isolation | Can flag suspicious behavior without a signature | Detects and contains affected endpoints | Helps hunt for persistence | Needs agents and tuning; weak on many appliances; best for servers and workstations |
XDR | Correlates endpoint, identity, email and network telemetry | Adds context to anomalies | Connects related alerts into one story | Helps scope the intrusion | Depends on integrated data sources; best where one vendor ecosystem is in use |
SIEM | Centralizes logs and correlation | Baselines and detection rules | Cross-source alerts and timelines | Retrospective searches | Only as good as logs and tuning; needs skilled staff |
MDR | Outsourced monitoring and response | Continuous watching if scoped | Human triage and containment | Support during recovery | Depends on contract scope and visibility; best for teams without 24/7 staff |
WAF | Filters HTTP requests to web applications | Generic protections | May block known exploit patterns as virtual patching | Can be relaxed once fixed | HTTP only; can be bypassed; best for exposed web apps as a stopgap |
IDS/IPS | Detects or blocks suspicious network traffic | Signatures for known issues | Anomaly rules may help | Blocks known n-day exploit traffic | Encryption limits visibility; signatures lag for zero-days |
Vulnerability management | Finds and prioritizes known weaknesses | Reduces n-day exposure | Prioritizes mitigation for new CVEs | Confirms fixes were deployed | Mostly known flaws; does not find unknown ones |
Attack-surface management | Discovers internet-exposed assets | Finds forgotten systems | Shows where a new flaw applies | Verifies removal of exposure | Discovery only; needs an owner for follow-up |
Penetration testing | Authorized testing of applications, APIs and configurations | Finds exploitable weaknesses in scope | Not a response tool | Retesting can verify fixes if scoped | Point-in-time and scoped; cannot guarantee finding all flaws or zero-days |
Incident response | Structured investigation and recovery | Planning and exercises | Containment and investigation | Verification and lessons learned | Needs preparation, authority and skilled people |
Backups | Restores data and systems | A recovery option | Limits ransomware damage | A rebuild source | Does not stop data theft; backups can be targeted; test restores |
Many teams assemble this from several products. Compare AI endpoint protection and AI SIEM offerings on integration and visibility, and review broader cybersecurity services options against the gaps in your own coverage.
How Penetration Testing Helps—and Its Limits
Penetration testing is authorized, scoped testing that simulates how an attacker might abuse your systems. Articsledge's penetration testing services cover professionally scoped engagements for web applications, APIs and SaaS platforms, with vulnerability assessments and remediation verification or retesting available where separately scoped. Engagements depend on agreed scope, technical arrangements, specialist availability and written authorization.
Testing helps with zero-day readiness in practical ways:
Finds weaknesses attackers could use first. Broken access control, weak session handling, risky configurations, exposed endpoints and business-logic flaws are common findings in applications and APIs.
Prioritizes remediation. A report ranks issues by exploitability and impact so limited engineering time goes to the right fixes.
Tests your assumptions. Segmentation, logging and alerting can be exercised so gaps show up before an incident does.
Verifies fixes. Retesting, where scoped, confirms that a change closed the weakness rather than moving it.
Its limits matter just as much. A test is a point-in-time exercise bounded by its scope. It cannot certify immunity to zero-days, prove that no vulnerabilities remain, or satisfy a compliance requirement on its own. Zero-days in third-party products such as a VPN appliance usually fall outside a typical application test. Testing complements patching, detection and response rather than replacing them. For deeper background, see the guide to penetration testing services, the primer on AI penetration testing, the overview of AI red teaming and the notes on cybersecurity consulting.
![]() SaaS Launch or Customer Security Review? Test Access Control and Tenant Isolation Customers often ask how you protect their data. Scoped testing of your SaaS application and APIs can examine access control, tenant isolation, authentication and session handling before a launch or a security questionnaire arrives. Articsledge arranges engagements after agreeing scope and written authorization. A report supports your evidence; it does not certify compliance or guarantee that no vulnerabilities exist. |
Choosing a Strategy by Organization Size
The right plan depends on size, staffing and what you protect. The table gives a minimum viable baseline for each group, not a complete program.
Organization | Minimum viable defenses | Trade-offs | Signals to look for a service |
Individuals | Automatic updates, unique passwords with passkeys or MFA, device encryption, backups | Low cost, limited visibility | Elevated-risk roles may want hardened device modes |
Small businesses | Asset list, managed updates, MFA, EDR on endpoints and servers, tested backups, a one-page response plan | Few staff, so managed services help | Nobody can watch alerts out of hours, or a customer-facing app handles sensitive data |
SaaS founders | Secure development practices, dependency tracking, logging, tenant-isolation checks, a WAF as a stopgap | Speed versus rigor | Customer security questionnaires, a launch, or a major architecture change |
Growing firms | Inventory and attack-surface management, SIEM with defined use cases, segmentation, exercises | Tool sprawl and unclear ownership | Cross-team incidents, audits or acquisitions |
Enterprises | Detection engineering, threat intelligence, edge-device governance, a recurring testing program, rehearsed response | Cost and complexity | Board reporting and regulatory obligations |
When you buy a service, look for written scope and authorization, clearly defined deliverables, an honest explanation of limits, and named options for retesting. Be wary of any provider that promises to find every vulnerability or to detect zero-days with certainty. Reviews of AI cybersecurity tools can help you compare detection products on evidence rather than slogans.
The Articsledge Zero-Day Readiness Decision Framework
This framework is an editorial decision aid created for this guide. It is not validated research, a standard or a forecast of whether a zero-day exists. It helps a team compare how urgently a specific system or newly disclosed flaw needs attention. Score each factor from 0 to 2, add the results, and adjust the factors to your own context.
Factor | 0 (lower concern) | 1 | 2 (higher concern) |
Exposure | Internal only | Reachable through VPN or partner link | Directly internet-facing |
Asset criticality | Low business impact | Moderate impact | Core revenue, safety or access path |
Privilege | Unprivileged and isolated | Some elevated access | Administrative or broad API access |
Sensitive data | None | Internal data | Customer, regulated or secret data |
Control coverage | Monitoring and segmentation in place | Partial coverage | Little or no monitoring, such as an appliance |
Exploitation intelligence | No reports | Proof of concept or vendor warning | Confirmed exploitation, such as a KEV listing |
Mitigation availability | Patch available and deployable | Workaround only | Nothing available yet |
Detection ability | Strong logging and alerts | Partial | Little visibility |
Recovery readiness | Tested rebuild and restore | Plans exist but are untested | No clear recovery path |
Interpreting the total (0 to 18). A score of 0 to 5 fits routine handling in the normal patch cycle. A score of 6 to 11 is elevated: schedule mitigation within days and add monitoring. A score of 12 to 18 is urgent: apply mitigations immediately, increase monitoring and prepare to isolate. Confirmed compromise overrides the score and calls for incident response.
Hypothetical worked example. A 40-person SaaS company runs an internet-facing VPN gateway. The vendor announces a flaw with confirmed exploitation and offers only a mitigation. The team scores exposure 2, criticality 2, privilege 2, sensitive data 1, control coverage 2 because no agent runs on the appliance, exploitation intelligence 2, mitigation availability 1, detection ability 1 and recovery readiness 1. The total is 14, which is urgent. The team applies the mitigation the same day, forwards the appliance logs, checks for indicators from the advisory, resets administrator credentials as a precaution and prepares a rebuild procedure for when the patch arrives.
Assumptions and limits. Factors carry equal weight, the scale is qualitative, and the example is invented for illustration. The framework was not tested against incident data. Teams with unusual risk profiles should reweight it, and they can place it inside the broader practices described in the guides to AI in cybersecurity and cybersecurity fundamentals.
Practical 30/60/90-Day Improvement Roadmap
A schedule helps when no incident is under way. Treat it as a starting point and adapt owners to your structure.
Window | Owner | Actions | Outputs | Checks |
Days 0 to 30 | IT lead and security owner | Inventory internet-facing assets, enable key logging, forward appliance logs, require MFA for administrators, verify backups, name an incident lead | Asset register, log source list, incident contact sheet | A test restore succeeds and every exposed asset has an owner |
Days 31 to 60 | Security and engineering | Segment critical systems, write patch triage rules using CVSS, EPSS, KEV and exposure, prepare mitigation playbooks for edge devices, run a tabletop exercise | Patch policy, playbooks, exercise notes | A dry run of disabling and isolating a service works, and time to mitigate is measured |
Days 61 to 90 | Leadership, security and engineering | Commission scoped testing of critical applications and APIs, fix and retest top findings, tune detections, review third-party exposure | Test report, remediation tracker, updated detection rules | Retesting closes priority findings and alerts are reviewed for false positives |
Important: if compromise is suspected, skip this schedule and begin incident response immediately.
![]() Ready to Define Scope, Objectives and a Quote? A useful quote starts with a clear scope: which applications, APIs and environments matter most, what you want to learn and what must stay off limits. Share those details with Articsledge to discuss objectives, authorization and timing for a scoped penetration test. Retesting and remediation support are separate scopes, and delivery depends on agreed arrangements and specialist availability. |
Frequently Asked Questions About Zero-Day Attacks
What is a zero-day attack in simple terms?
A zero-day attack exploits a software or hardware flaw before the vendor has a fix, or before defenders knew the flaw existed. The attacker has a head start because there is no patch to apply, so organizations rely on mitigations, monitoring and containment until a fix arrives.
What is the difference between a zero-day vulnerability, exploit and attack?
The vulnerability is the weakness, the exploit is the technique or code that abuses it, and the attack is its actual use against a target. A flaw can exist for years without being exploited, and an exploit can exist without ever being used against you.
How often do zero-day attacks happen?
Google Threat Intelligence Group counted 90 exploited zero-days in 2025, 78 in 2024 and 100 in 2023. These are detected and disclosed vulnerabilities, not victims, counts can change as older incidents surface, and GTIG notes its data may undercount total activity.
Can antivirus or EDR stop a zero-day attack?
Sometimes, but not reliably. Behavior-based EDR can flag suspicious actions without a signature, yet it needs agents, tuning and monitoring, and many appliances cannot run one. Treat endpoint tools as one layer alongside patching, segmentation, logging and response, not as a guarantee.
Does patching protect against zero-days?
Patching removes the exposure once a fix exists, but it cannot help before disclosure and does not remove an intruder already inside. After patching, review logs, reset credentials and hunt for persistence, as the Ivanti advisories illustrate.
How is a zero-day different from a zero-click attack?
Zero-day describes patch status and prior knowledge, while zero-click describes whether the victim must act. The BLASTPASS iMessage chain was both. An n-day exploit can also be zero-click, and a zero-day can require a click.
Can a vulnerability scanner find zero-days?
Scanners mainly identify known weaknesses such as outdated versions and common misconfigurations, so they generally cannot find undisclosed flaws. They remain valuable for removing n-day exposure and for showing where a newly disclosed flaw would apply across your estate.
Can a penetration test find a zero-day?
Rarely. A scoped test focuses on your applications, APIs and configuration, and may uncover previously unknown flaws in your own code, but it is not a zero-day research program and cannot guarantee discovery of unknown flaws in third-party products.
Are CVSS and EPSS useful for zero-days?
CVSS measures severity characteristics, not likelihood of exploitation. EPSS estimates the chance that a published CVE will be exploited within 30 days. Neither describes undisclosed flaws, so use them to prioritize known issues and add exposure and exploitation intelligence such as CISA KEV.
What should small businesses prioritize first?
Start with an inventory, automatic updates, MFA for administrators and remote access, tested backups and basic logging. Add EDR on endpoints and servers and a one-page response plan. Consider managed monitoring if nobody can watch alerts out of hours.
What should I do if I think a zero-day was used against me?
Treat it as an incident. Preserve evidence, isolate affected systems, follow vendor advisories, reset credentials and sessions, hunt for persistence and involve legal counsel. Engage an incident-response specialist rather than waiting for a scheduled security review.
Does Articsledge provide incident response or zero-day research?
Articsledge's service page describes scoped penetration testing for web applications, APIs and SaaS platforms, with assessments and retesting separately scoped. It does not present itself as an incident-response, SOC or zero-day research provider, so use a specialist for active incidents.
Key Takeaways
Zero-day describes timing and knowledge, not sophistication. Always ask when the flaw became public and whether a fix existed.
Edge devices and business software stay in the spotlight. GTIG found enterprise technology at a record 48% share of 2025 zero-days, with edge devices prominent.
Precise counting prevents bad conclusions. Disclosed vulnerabilities, exploited CVEs, campaigns and victims are different numbers over different windows.
Mitigation is a skill to rehearse. In several documented cases, vendors offered mitigations before patches, and teams that had practiced moved faster.
Post-patch investigation is part of the fix. Credentials, sessions and persistence outlive a software update.
Controls cover different moments. Use the matrix to find where you have nothing before disclosure, during exploitation or after a patch.
Testing finds weaknesses in scope, not certainty. It strengthens application and API defenses but cannot certify safety against unknown flaws.
Transparent decision aids beat gut feel. Whatever scoring you use, write down the assumptions and revisit them after each major disclosure.
Actionable Next Steps
This week, security owner: list every internet-facing system with its owner and version.
This week, IT lead: confirm automatic updates, administrator MFA and a tested backup restore.
Within two weeks, security owner: forward logs from edge devices and applications to central storage with enough retention to investigate older activity.
Within 30 days, leadership: name an incident lead, an escalation path and the external specialists you would call.
Within 45 days, engineering: write a patch triage rule that combines CVSS, EPSS, KEV and exposure, plus mitigation playbooks for edge devices.
Within 60 days, security and business owners: run a tabletop exercise for an edge-device zero-day.
Within 90 days, product and security owners: scope authorized testing for critical applications and APIs, and if useful discuss a scoped assessment with Articsledge.
Ongoing, security owner: review framework scores quarterly and after every major disclosure.
Glossary
Vulnerability: a weakness in software, firmware or hardware that could be abused.
Exploit: a technique or code that takes advantage of a vulnerability.
Zero-day attack: the use of an exploit before a fix exists or before the flaw was known to the vendor.
N-day: a flaw that already has a public fix or disclosure, exploited after the fact.
Zero-click: an exploit that needs no action from the victim.
CVE: Common Vulnerabilities and Exposures, a public identifier for a disclosed vulnerability.
CVSS: Common Vulnerability Scoring System, which rates severity characteristics, not exploitation likelihood.
EPSS: Exploit Prediction Scoring System, a probability that a published CVE will see exploitation within 30 days.
KEV: CISA's Known Exploited Vulnerabilities catalog, which lists flaws with confirmed exploitation, not only zero-days.
Exploit chain: several flaws used in sequence, for example to run code and then escape a sandbox.
EDR and XDR: endpoint detection and response, and its extended form that correlates more data sources.
WAF: web application firewall, which filters HTTP traffic to web applications.
Virtual patching: using a filter or rule to block exploitation of a flaw without changing the vulnerable software.
Penetration testing: authorized, scoped testing that simulates attacker behavior to find exploitable weaknesses.
Retesting: checking after remediation that a reported weakness is actually fixed.
Persistence: methods an intruder uses to keep access after reboots, patches or password changes.
Sources & References
NIST Computer Security Resource Center Glossary, "Zero Day Attack". NIST. Current entry.
Google Threat Intelligence Group, "2025 Zero-Day Review". Google Cloud Blog. March 5, 2026.
Google Threat Intelligence Group, "Vulnerability Discovery and Exploitation Trends in the AI Era". Google Cloud Blog. September 30, 2026.
CISA, "Threat Actors Exploit MOVEit Transfer Vulnerability," advisory AA23-158A. Cybersecurity and Infrastructure Security Agency. Updated June 16, 2023.
CISA, Emergency Directive 24-01 and supplemental directions on Ivanti Connect Secure and Policy Secure. CISA. Supplemental directions January 31 and February 9, 2024.
Mandiant, "Cutting Edge: Suspected APT Targets Ivanti Connect Secure VPN in New Zero-Day Exploitation". Google Cloud Blog. January 11, 2024.
Microsoft Threat Intelligence, "HAFNIUM targeting Exchange Servers with 0-day exploits". Microsoft Security Blog. March 2, 2021.
Citizen Lab, "BLASTPASS: NSO Group iPhone Zero-Click, Zero-Day Exploit Captured in the Wild". University of Toronto. September 7, 2023.
Kaspersky GReAT, "Operation ForumTroll: APT attack with Google Chrome zero-day exploit chain". Securelist. March 25, 2025.
Oracle, "Security Alert CVE-2025-61882". Oracle. October 4, 2025.
Falliere, O Murchu and Chien, "W32.Stuxnet Dossier". Symantec Security Response. Version 1.4, February 2011.
FIRST, "EPSS Frequently Asked Questions". Forum of Incident Response and Security Teams. Current page.
NIST, SP 800-61 Rev. 3, "Incident Response Recommendations and Considerations for Cybersecurity Risk Management: A CSF 2.0 Community Profile". NIST. April 2025.
CISA, Known Exploited Vulnerabilities Catalog. CISA. Continuously updated.
MITRE ATT&CK, "Exploit Public-Facing Application (T1190)". MITRE. Current entry.
OWASP, Web Security Testing Guide. OWASP Foundation. Current project page.
Articsledge, Penetration Testing Services. Articsledge. Reviewed for scope on November 23, 2026.







