Back to the trend of the day: Some hackers use AI agents to turn public software fixes into attacks
Poly Rating · from the trend of the day

Patch to Exploit Time 2026: 10 Products Hit Fastest

Attackers using AI agents built working PaperCut exploits within a day of its August 2026 fixes and hit 395 organisations about four days later, while Mandiant now puts the average time-to-exploit at minus 7 days.

  • No. 1: PaperCut NG/MF
  • 10 places

01Final list

The leading AIs each built their own list of ten independently. Here is one result that is easy to check: every place has a reason, a source, and the list of AIs that included it.

Days from patch release to the first confirmed exploitation in the wild, using Mandiant's time-to-exploit data

A bright badge means this AI put the entry on its list. A dim badge means it did not.

  1. 1

    PaperCut NG/MF

    Exploits were working within a day of the 27 Aug 2026 fix, and an AI-agent campaign hit 395 organisations about four days later. One reading calls the chain a zero-day pair.

    Included it: Meta, Grok, MiMo, Kimi. Did not include: Claude, ChatGPT, Nemotron, Gemini, DeepSeek, Wenxin.

    In their lists: Meta: 4 · Grok: 1 · MiMo: 1 · Kimi: 10

    What each AI noted 4
    • Metaplace 4

      Working exploits within a day of disclosure, and a likely LLM-built campaign against hundreds of organisations four days later.

      Source ↗
    • Grokplace 1

      AI agents turned two August 2026 patches into RCE exploits in under a day and a 395-org campaign in four days

    • MiMoplace 1

      A fix on 27 August, then an AI-agent swarm through 440 print servers at 395 organizations in 48 countries within days — the fastest documented fix-to-weapon cycle of the 24 months.

      Source ↗
    • Kimiplace 10

      The case Google calls a first: an AI agent pipeline turned a published fix into a working attack with no human help.

      Source ↗
    4 AIs put it here
  2. 2

    Citrix NetScaler ADC / Gateway

    CVE-2026-8451 was exploited within 24 hours of the 1 July 2026 patch. CitrixBleed 2 showed exploitation evidence before its 2025 patch was two weeks old.

    Included it: Claude, Grok, Gemini, MiMo, Kimi. Did not include: ChatGPT, Meta, Nemotron, DeepSeek, Wenxin.

    In their lists: Claude: 2 · Grok: 3 · Gemini: 2 · MiMo: 7 · Kimi: 7

    What each AI noted 5
    • Claudeplace 2

      Exploited within 24 hours of the July 1 patch; earlier CitrixBleed variants were also hit within days.

      Source ↗
    • Grokplace 3

      Repeated 2026 zero-days (including ToolShell-era patterns) entered KEV within days; edge appliance focus keeps windows short

    • Geminiplace 2

      Memory-leak and buffer-overflow fixes were weaponized into public session-hijacking scripts in under 24 hours.

      Source ↗
    • MiMoplace 7

      A perennial KEV fixture: when its fixes land, the exploit evidence historically arrives in days, not weeks.

      Source ↗
    • Kimiplace 7

      Evidence of session-hijacking exploitation surfaced before the patch was two weeks old, and ransomware crews were still chaining it a year later.

      Source ↗
    5 AIs put it here
  3. 3

    Ivanti Connect Secure

    Exploited about three weeks before its January 2025 patch, and attackers turned a fix Ivanti rated low-risk into working RCE. Ivanti's median time from CVE to KEV is 0.8 days.

    Included it: Claude, Grok, Gemini, MiMo, Kimi, DeepSeek. Did not include: ChatGPT, Meta, Nemotron, Wenxin.

    In their lists: Claude: 9 · Grok: 2 · Gemini: 1 · MiMo: 4 · Kimi: 3 · DeepSeek: 1

    What each AI noted 6
    • Claudeplace 9

      Exploitation surged after attacks were traced back to July 2025, so some activity predated any fix.

    • Grokplace 2

      Multiple 2026 zero-day chains reached root without shells; CISA KEV listings within days of disclosure

    • Geminiplace 1

      Averaged -14 to 0 days from patch to attack, with adversary exploitation consistently starting before or within 12 hours of advisory drops.

    • MiMoplace 4

      21 of Google's 90 zero-days found in 2025 sat in security and networking gear — the products meant to protect you are the favorite target.

    • Kimiplace 3

      Attackers studied a patch Ivanti rated low-risk and turned it into working remote code execution — the fix itself became the exploit manual.

      Source ↗
    • DeepSeekplace 1

      Multiple 2026 CVEs were exploited on the same day they became public — and the vendor’s median KEV lag is 0.8 days.

      Source ↗
    6 AIs put it here
  4. 4

    Microsoft SharePoint Server

    ToolShell was exploited from 17–18 July 2025, before the emergency patch, and bypassed the fixes shipped on 8 July. Microsoft's median time from CVE to KEV is −0.7 days.

    Included it: Claude, Meta, Grok, MiMo, Kimi, DeepSeek, Wenxin. Did not include: ChatGPT, Nemotron, Gemini.

    In their lists: Claude: 6 · Meta: 8 · Grok: 4 · MiMo: 5 · Kimi: 6 · DeepSeek: 3 · Wenxin: 8

    What each AI noted 7
    • Claudeplace 6

      39 KEV additions in 2025, more than any other vendor, and one 2026 flaw was attacked in under 48 hours.

      Source ↗
    • Metaplace 8

      Microsoft says exploitation attempts began 7 July 2025 — a day before the first ToolShell patches — then a patch bypass exploded at scale on 18 July.

      Source ↗
    • Grokplace 4

      Highest absolute KEV volume; several 2026 flaws moved from patch to confirmed exploitation in days to weeks

    • MiMoplace 5

      Exploits stayed the No. 1 way in, and most initial-access exploits Mandiant saw were zero-days — the biggest installed base runs the shortest clock.

      Source ↗
    • Kimiplace 6

      Exploited two days before the emergency patch — and the exploit chain bypassed the patches Microsoft had shipped eleven days earlier.

      Source ↗
    • DeepSeekplace 3

      Microsoft’s overall median KEV lag is negative (−0.7 days), and SharePoint flaws have been exploited before or within days of patch Tuesday.

      Source ↗
    • Wenxinplace 8

      Code injection RCE flaw added to KEV less than 7 days after public patch release

      Source ↗
    7 AIs put it here
  5. 5

    WordPress (core and plugins)

    Heavily targeted plugin flaws reach mass exploitation in a median of 5 hours from disclosure. Core flaws ranged from within hours (CVE-2026-87902) to 3 days to KEV (CVE-2026-119977).

    Included it: Claude, Meta, MiMo, DeepSeek, Wenxin. Did not include: ChatGPT, Grok, Nemotron, Gemini, Kimi.

    In their lists: Claude: 5 · Meta: 1 · MiMo: 6 · DeepSeek: 5 · Wenxin: 6

    What each AI noted 5
    • Claudeplace 5

      A critical flaw was exploited within hours of its patch on the world's most common CMS.

      Source ↗
    • Metaplace 1

      Heavily targeted plugin flaws reach mass exploitation in a median of 5 hours — faster than most teams can approve an emergency change.

      Source ↗
    • MiMoplace 6

      exploit_agent reportedly examined a WordPress plugin as its study material — ordinary website add-ons are now the practice range for autonomous exploit builders.

      Source ↗
    • DeepSeekplace 5

      CVE-2026-119977 was exploited in 3 days, faster than most enterprise products’ entire patch cycle.

    • Wenxinplace 6

      LFI/RCE flaw added to KEV 3 days after the official patch was published

      Source ↗
    5 AIs put it here
  6. 6

    GitLab CE / EE

    CVE-2026-85706 was exploited within a day of its 10 Sep 2026 patch and listed on KEV on 11 Sep. Sources disagree on CVE-2026-97229: under a day or 17 days to KEV.

    Included it: Meta, DeepSeek. Did not include: Claude, ChatGPT, Grok, Nemotron, Gemini, MiMo, Kimi, Wenxin.

    In their lists: Meta: 3 · DeepSeek: 7

    What each AI noted 2
    • Metaplace 3

      CVE-2026-85706 was exploited within a day of its patch and landed in CISA KEV the next day.

    • DeepSeekplace 7

      CVE-2026-97229 was exploited on the day it was published, with a PoC available within hours.

    2 AIs put it here
  7. 7

    Palo Alto Networks PAN-OS

    CVE-2025-0108: the patch and a technical write-up came out on 12 Feb 2025, and GreyNoise saw attempts the next day. Another timeline dates confirmed active exploitation to 18 Feb.

    Included it: Meta, Gemini, Kimi. Did not include: Claude, ChatGPT, Grok, Nemotron, MiMo, DeepSeek, Wenxin.

    In their lists: Meta: 5 · Gemini: 3 · Kimi: 9

    What each AI noted 3
    • Metaplace 5

      A public PoC appeared the day after Palo Alto's bulletin, and GreyNoise saw active exploitation six days after the bulletin.

      Source ↗
    • Geminiplace 3

      Firewall management interfaces saw zero-click root exploit chains weaponized in under 48 hours following security bulletins.

      Source ↗
    • Kimiplace 9

      The fastest clean conversion on record: patch and advisory on February 12, exploitation attempts observed February 13 — one day.

      Source ↗
    3 AIs put it here
  8. 8

    BeyondTrust Remote Support

    CVE-2026-102764 was attacked from 29 Aug 2026, about 40 days before the 8 Oct patch, and reached CISA KEV within a day of disclosure.

    Included it: Claude, Kimi, DeepSeek. Did not include: ChatGPT, Meta, Grok, Nemotron, Gemini, MiMo, Wenxin.

    In their lists: Claude: 4 · Kimi: 2 · DeepSeek: 10

    What each AI noted 3
    • Claudeplace 4

      A remote-code-execution flaw attacked hours after the patch shipped, in a tool that guards privileged access.

      Source ↗
    • Kimiplace 2

      Seven weeks of attacks against a security vendor's own remote-support appliance before anyone could patch.

      Source ↗
    • DeepSeekplace 10

      CVE-2026-102764 was added to KEV on 16 October 2026, the day after disclosure, with a PoC already referenced.

    3 AIs put it here
  9. 9

    SAP NetWeaver Visual Composer

    CVE-2025-31324 (CVSS 10.0) was exploited before SAP's 24 April 2025 emergency patch, and Mandiant named it the most exploited flaw of 2025. SAP at first disputed the reports.

    Included it: Grok, Kimi. Did not include: Claude, ChatGPT, Meta, Nemotron, Gemini, MiMo, DeepSeek, Wenxin.

    In their lists: Grok: 8 · Kimi: 4

    What each AI noted 2
    • Grokplace 8

      Top 2025 initial-access vector per Mandiant; patch-to-exploitation windows measured in days for key CVEs

    • Kimiplace 4

      The most-exploited vulnerability of Mandiant's entire 2025 report was being used in attacks before SAP shipped the fix.

      Source ↗
    2 AIs put it here
  10. 10

    SonicWall SSL VPN / SonicOS

    CVE-2024-53704 was exploited within days of its Feb 2025 technical write-up. CVE-2026-19868 went from publication to KEV in 0 days, against a 124-day median for the product.

    Included it: Meta, DeepSeek. Did not include: Claude, ChatGPT, Grok, Nemotron, Gemini, MiMo, Kimi, Wenxin.

    In their lists: Meta: 6 · DeepSeek: 2

    What each AI noted 2
    • Metaplace 6

      Full technical details on 10 February 2025, exploitation within days, and CISA KEV eight days later.

      Source ↗
    • DeepSeekplace 2

      CVE-2026-19868 went from disclosure to confirmed exploitation in 0 days, against a 124-day median for the product.

    2 AIs put it here

02Worth knowing

What the AIs noticed while building the list.

Mandiant's mean time-to-exploit fell from 63 days in 2018–19 to an estimated −7 days in 2025, so on average exploitation now starts before the fix exists.

Noticed by:
Included it: Claude, Grok, Gemini, MiMo, Kimi, DeepSeek. Did not include: ChatGPT, Meta, Nemotron, Wenxin.
Source ↗

VulnCheck found 28.96% of 2025 KEVs exploited on or before CVE publication (28.93% in its later PDF). That fell to 23.43% in 1H 2026, while the median time from CVE to KEV dropped from 120 to 80 days.

Noticed by:
Included it: Claude, Meta, Grok, Kimi, DeepSeek. Did not include: ChatGPT, Nemotron, Gemini, MiMo, Wenxin.
Source ↗

The clocks disagree by design. Root Evidence finds a median of 116 days from disclosure to first exploitation for CVEs first exploited in 2026, against Mandiant's −7-day mean measured from the patch.

Noticed by:
Included it: Grok, MiMo, Kimi. Did not include: Claude, ChatGPT, Meta, Nemotron, Gemini, DeepSeek, Wenxin.
Source ↗

A KEV date is a filing date: WhatsUp Gold CVE-2024-4885 was listed 234 days after VulnCheck's first exploit evidence.

Noticed by:
Included it: MiMo. Did not include: Claude, ChatGPT, Meta, Grok, Nemotron, Gemini, Kimi, DeepSeek, Wenxin.
Source ↗

Exploited flaws per month rose from 10.5 in 2025 to 18 in January–August 2026, mostly through fast weaponization of already-patched n-days, though only 0.23% of 2026's disclosed flaws were seen exploited.

Noticed by:
Included it: Kimi. Did not include: Claude, ChatGPT, Meta, Grok, Nemotron, Gemini, MiMo, DeepSeek, Wenxin.
Source ↗

03Questions people ask

What is patch-to-exploit time?

It is the number of days between a fix being released and the first exploitation in the wild. Mandiant's 2025 mean is an estimated −7 days, meaning attacks usually begin before the patch.

Source ↗
How fast are published patches turned into attacks?

In the fastest 2026 cases it takes about a day: PaperCut exploits worked within a day of disclosure, and a campaign against 395 organisations followed about four days later.

Source ↗
How fast do pimple patches work?

Hydrocolloid pimple patches usually show visible results in about 6–8 hours, which is why they are worn overnight. That is unrelated to software patches, which attackers can now weaponize within a day.

Source ↗
Why do patch-to-exploit numbers differ between sources?

Mandiant counts from the patch, VulnCheck from CVE publication, and CISA records the listing date. That is how one study gets a −7-day mean and another a 116-day median.

Source ↗

04How we counted

In plain words, no formulas.

1The question

Top 10 Software Products Where a Published Fix Turns Into an Attack Fastest

2Why this topic

Mandiant times the race from the patch and VulnCheck from the CVE, so the two clocks disagree. Setting them against CISA's confirmed list shows which products are attacked before defenders can even start, and no single table shows that.

3Where it applies

Global; widely used software products with publicly patched flaws, last 24 months

4What we looked at

  • Days from patch release to the first confirmed exploitation in the wild, using Mandiant's time-to-exploit data
  • Days from CVE publication to the first public exploit evidence, using VulnCheck KEV
  • Share of each product's disclosed flaws that end up on CISA's confirmed-exploited list
  • Gap between the two clocks: a product counts as more urgent when exploitation starts before or close to the fix
  • Google Threat Intelligence / Mandiant (M-Trends 2026) — Mean days between patch release and first exploitation seen in incident-response cases; the 2025 estimate is −7 days
  • VulnCheck State of Exploitation 1H 2026 — Days from CVE publication to exploitation evidence, and the share of KEVs exploited on or before publication day
  • CISA Known Exploited Vulnerabilities Catalog — CVEs confirmed exploited in the wild, by vendor and product, with the date listed (not the date exploitation began)
  • CVETodo vendor rankings — Median days from CVE publication to CISA KEV listing, per vendor

5Who ranked

The leading AIs, each building its own list of ten independently and naming a source for every place: Claude, ChatGPT, Meta, Grok, Nemotron, Gemini, MiMo, Kimi, DeepSeek, Wenxin.

6How the places were set

The same entry written differently by different AIs was counted as one. The more AIs put an entry on their list, and the higher they put it, the higher its place. "AI agreement" shows how many of the AIs had it on their own list. Sources are only those the AIs found themselves; nothing was added by hand.

05Each AI's own list

🇺🇸 Claude10 places · 2 notes

The same question, answered by Claude alone. The final list above combines all the AIs.

  1. 1
    Adobe ColdFusion (2026)

    About two hours from the June 30 patch to the first honeypot hit, the fastest case I found.

    Source ↗
  2. 2
    Citrix NetScaler (2026)

    Exploited within 24 hours of the July 1 patch; earlier CitrixBleed variants were also hit within days.

    Source ↗
  3. 3
    cPanel (2026)

    Several threat actors were seen within 24 hours, and write-ups from patch diffing followed within days.

    Source ↗
  4. 4
    BeyondTrust remote access (2026)

    A remote-code-execution flaw attacked hours after the patch shipped, in a tool that guards privileged access.

    Source ↗
  5. 5
    WordPress (critical flaw, 2026)

    A critical flaw was exploited within hours of its patch on the world's most common CMS.

    Source ↗
  6. 6
    Microsoft Windows, Office and SharePoint

    39 KEV additions in 2025, more than any other vendor, and one 2026 flaw was attacked in under 48 hours.

    Source ↗
  7. 7
    ServiceNow AI Platform (2026)

    Four days from patch to confirmed exploitation.

    Source ↗
  8. 8
    Weaver E-cology (2026)

    Attacks began within five days of the vendor patch.

    Source ↗
  9. 9
    Ivanti (Sentry, EPMM and others)

    Exploitation surged after attacks were traced back to July 2025, so some activity predated any fix.

  10. 10
    Fortinet

    8 KEV additions in 2025, with a 3-day federal deadline on the April 2026 wave.

    Source ↗

What Claude noticed

Mandiant's mean time-to-exploit for 2025 is estimated at -7 days. On average, attackers start before the fix exists.

Google says most exploitation growth comes from n-days (flaws already disclosed). It suggests attackers may use LLMs to analyze patches and proof-of-concept code, though it hedges on how widely.

🇺🇸 Meta10 places · 4 notes

The same question, answered by Meta alone. The final list above combines all the AIs.

  1. 1
    WordPress plugins

    Heavily targeted plugin flaws reach mass exploitation in a median of 5 hours — faster than most teams can approve an emergency change.

    Source ↗
  2. 2
    Adobe Commerce / Magento Open Source

    BlockStrap was reproduced four ways within 24 hours of Adobe's patch, and StyleSmuggler was already exploited before a fix existed.

    Source ↗
  3. 3
    GitLab CE / EE

    CVE-2026-85706 was exploited within a day of its patch and landed in CISA KEV the next day.

  4. 4
    PaperCut print management servers

    Working exploits within a day of disclosure, and a likely LLM-built campaign against hundreds of organisations four days later.

    Source ↗
  5. 5
    Palo Alto Networks PAN-OS

    A public PoC appeared the day after Palo Alto's bulletin, and GreyNoise saw active exploitation six days after the bulletin.

    Source ↗
  6. 6
    SonicWall SSL VPN / SonicOS

    Full technical details on 10 February 2025, exploitation within days, and CISA KEV eight days later.

    Source ↗
  7. 7
    Fortinet products

    CVE-2025-32756 was exploited as a zero-day, so the patch arrived after exploitation had already started.

  8. 8
    Microsoft SharePoint Server (ToolShell)

    Microsoft says exploitation attempts began 7 July 2025 — a day before the first ToolShell patches — then a patch bypass exploded at scale on 18 July.

    Source ↗
  9. 9
    VMware ESXi

    Attackers had more than a year of exploitation before the patch — the longest negative gap in this ranking.

  10. 10
    Apache Log4j

    The classic warning: exploitation was seen 1 December 2021, eight days before public disclosure, and mass scanning followed within hours.

    Source ↗

What Meta noticed

Google's Q4 2026 tracker says exploit_agent is the first known case of an AI agent turning a fix into a working attack with no human help — it does continuous differential analysis of popular web software, drafts exploits and executes them.

PaperCut shows the new compression: working exploits within a day of disclosure, then a likely LLM-built campaign against hundreds of organisations four days later, reaching domain admin within minutes.

VulnCheck's 2025 data: 28.96% of known-exploited flaws showed exploitation on or before the day the CVE was published — up from earlier years — so the patch clock often starts late.

The fastest measured window is not a firewall or server at all: heavily targeted WordPress plugin flaws hit mass exploitation in a median of 5 hours.

🇺🇸 Grok10 places

The same question, answered by Grok alone. The final list above combines all the AIs.

  1. 1
    PaperCut NG/MF (print management)

    AI agents turned two August 2026 patches into RCE exploits in under a day and a 395-org campaign in four days

  2. 2
    Ivanti Connect Secure / Pulse (VPN gateway)

    Multiple 2026 zero-day chains reached root without shells; CISA KEV listings within days of disclosure

  3. 3
    Citrix NetScaler ADC / Gateway

    Repeated 2026 zero-days (including ToolShell-era patterns) entered KEV within days; edge appliance focus keeps windows short

  4. 4
    Microsoft SharePoint / Office / Windows

    Highest absolute KEV volume; several 2026 flaws moved from patch to confirmed exploitation in days to weeks

  5. 5
    Cisco (IOS / ASA / SD-WAN / BroadWorks)

    Second-highest 2026 KEV pace; edge and management flaws routinely confirmed exploited within weeks

  6. 6
    Fortinet (FortiMail / FortiOS / multiple)

    Consistent high KEV density for edge appliances; 2026 path-traversal and related flaws added quickly after wild use

  7. 7
    Oracle (EBS / PeopleSoft / WebLogic)

    High-profile 2025-2026 initial-access flaws (including ShinyHunters campaigns) showed short post-patch windows

  8. 8
    SAP NetWeaver

    Top 2025 initial-access vector per Mandiant; patch-to-exploitation windows measured in days for key CVEs

  9. 9
    Google Chromium / Chrome / V8

    Frequent KEV additions with near-zero lag on some 2026 entries; browser engine speed keeps exploit clocks short

  10. 10
    Linux Kernel

    Steady KEV flow; local privilege-escalation flaws often reach confirmed exploitation before broad patch uptake

🇺🇸 Gemini10 places · 3 notes

The same question, answered by Gemini alone. The final list above combines all the AIs.

  1. 1
    Ivanti Connect Secure (ICS)

    Averaged -14 to 0 days from patch to attack, with adversary exploitation consistently starting before or within 12 hours of advisory drops.

  2. 2
    Citrix NetScaler ADC & Gateway

    Memory-leak and buffer-overflow fixes were weaponized into public session-hijacking scripts in under 24 hours.

    Source ↗
  3. 3
    Palo Alto Networks PAN-OS

    Firewall management interfaces saw zero-click root exploit chains weaponized in under 48 hours following security bulletins.

    Source ↗
  4. 4
    Fortinet FortiOS

    SSL-VPN and administrative web portal patches were reverse-engineered within 2 days to bypass authentication filters.

  5. 5
    Microsoft Exchange Server

    Cumulative updates regularly yielded operational remote code execution exploits via patch diffing in 2 to 4 days.

    Source ↗
  6. 6
    Atlassian Confluence Server & Data Center

    Template and OGNL injection patches collapsed into active mass-exploitation campaigns in under 72 hours.

    Source ↗
  7. 7
    VMware ESXi & vCenter

    Hypervisor management fixes triggered ransomware-focused exploitation scripts within 4 days of patch publishing.

  8. 8
    Apache ActiveMQ

    OpenWire serialization fixes were reversed and bundled into Kinsing and TellYouThePass attacks in just 48 hours.

    Source ↗
  9. 9
    F5 BIG-IP

    Management GUI and iControl REST updates were weaponized into unauthenticated command-injection attacks within 3 days.

    Source ↗
  10. 10
    Cisco IOS XE

    Web UI implant vulnerabilities saw mass device compromise begin within 24 hours of advisory issuance.

    Source ↗

What Gemini noticed

The overall mean time-to-exploit across major enterprise incidents plunged to -7 days in Mandiant telemetry because threat actors weaponize bugs as zero-days before software vendors can release a fix.

Network perimeter and edge devices represent over 70% of vulnerabilities exploited within 48 hours of patch publication due to privileged access and lack of host-based endpoint security agents.

Threat actors employ automated continuous patch-diffing pipelines that analyze GitHub commits and vendor firmware releases to generate proof-of-concept exploits in under 15 minutes.

🇨🇳 MiMo10 places · 1 notes

The same question, answered by MiMo alone. The final list above combines all the AIs.

  1. 1
    PaperCut NG/MF (2026)

    A fix on 27 August, then an AI-agent swarm through 440 print servers at 395 organizations in 48 countries within days — the fastest documented fix-to-weapon cycle of the 24 months.

    Source ↗
  2. 2
    Progress WhatsUp Gold (2024–2025)

    The catalog clock ran 234 days behind the attack clock: exploit evidence existed for CVE-2024-4885 nearly eight months before CISA listed it.

  3. 3
    Barracuda products (2026)

    Sitting on CISA's KEV front page right now — CVE-2026-111560, an OS command injection in multiple products, is today's reminder that edge appliances keep topping the list.

    Source ↗
  4. 4
    Security and networking appliances (FortiOS, Ivanti, PAN-OS class)

    21 of Google's 90 zero-days found in 2025 sat in security and networking gear — the products meant to protect you are the favorite target.

  5. 5
    Microsoft edge and Windows services

    Exploits stayed the No. 1 way in, and most initial-access exploits Mandiant saw were zero-days — the biggest installed base runs the shortest clock.

    Source ↗
  6. 6
    Web software and CMS plugins (WordPress-class)

    exploit_agent reportedly examined a WordPress plugin as its study material — ordinary website add-ons are now the practice range for autonomous exploit builders.

    Source ↗
  7. 7
    Citrix NetScaler (class)

    A perennial KEV fixture: when its fixes land, the exploit evidence historically arrives in days, not weeks.

    Source ↗
  8. 8
    Cisco IOS and security appliances

    One of the two vendors that populate every KEV wave — the catalog clock for its bugs tends to be days old before small IT teams even read the advisory.

    Source ↗
  9. 9
    VMware and virtualization platforms

    Hypervisor flaws are the rare class where exploitation can out the patch itself — a minus-day clock by construction.

  10. 10
    Atlassian Confluence and collaboration servers

    Every new CVE on these lands in the same funnel: public exploit code within a day, KEV listing later — 28.3% of KEVs show that shape.

    Source ↗

What MiMo noticed

28.3% of known-exploited flaws had public exploitation evidence within one day of the CVE going live — the disclosure itself is the trigger, not the patch.

🇨🇳 Kimi10 places · 3 notes

The same question, answered by Kimi alone. The final list above combines all the AIs.

  1. 1
    Oracle E-Business Suite (CVE-2025-61882, Cl0p campaign)

    Cl0p was exploiting it for about eight weeks before the emergency patch existed — the deepest negative clock of the period.

    Source ↗
  2. 2
    BeyondTrust Remote Support (CVE-2026-102764, Armory Castle)

    Seven weeks of attacks against a security vendor's own remote-support appliance before anyone could patch.

    Source ↗
  3. 3
    Ivanti Connect Secure (CVE-2025-0282, CVE-2025-22457)

    Attackers studied a patch Ivanti rated low-risk and turned it into working remote code execution — the fix itself became the exploit manual.

    Source ↗
  4. 4
    SAP NetWeaver Visual Composer (CVE-2025-31324)

    The most-exploited vulnerability of Mandiant's entire 2025 report was being used in attacks before SAP shipped the fix.

    Source ↗
  5. 5
    WinRAR (CVE-2025-8088, RomCom)

    A desktop archiver, exploited by RomCom 12 days before version 7.13 fixed it — and still exploited as an n-day by state actors months later.

    Source ↗
  6. 6
    Microsoft SharePoint Server (ToolShell, CVE-2025-53770/53771)

    Exploited two days before the emergency patch — and the exploit chain bypassed the patches Microsoft had shipped eleven days earlier.

    Source ↗
  7. 7
    Citrix NetScaler ADC/Gateway (Citrix Bleed 2, CVE-2025-5777)

    Evidence of session-hijacking exploitation surfaced before the patch was two weeks old, and ransomware crews were still chaining it a year later.

    Source ↗
  8. 8
    Metabase (CVE-2026-72898)

    Attacks hit Metabase Cloud on August 3; the advisory and patches arrived August 6 — three days of attack with nothing to install.

    Source ↗
  9. 9
    Palo Alto Networks PAN-OS (CVE-2025-0108)

    The fastest clean conversion on record: patch and advisory on February 12, exploitation attempts observed February 13 — one day.

    Source ↗
  10. 10
    PaperCut NG/MF (September 2026 AI-agent patch reversal)

    The case Google calls a first: an AI agent pipeline turned a published fix into a working attack with no human help.

    Source ↗

What Kimi noticed

Mandiant's mean time-to-exploit for 2025 is an estimated minus 7 days — exploitation now routinely starts a week before the patch exists, the first negative reading in the series (63 days in 2018-19).

For 2026 CVEs confirmed exploited, the median time from CVE publication to CISA KEV listing was 4 days, and 27% were listed the same day or earlier — about half of those same-day cases were network and security appliances (Cisco, Citrix, Ivanti, Palo Alto, Fortinet, F5, SonicWall).

The exploited share is tiny but the pace doubled: Google tracked only 0.23% of 2026's disclosed vulnerabilities as exploited, yet monthly exploitation jumped from a 10.5 average in 2025 to 18 in 2026 — concentrated in weaponizing already-patched n-days.

🇨🇳 DeepSeek10 places · 3 notes

The same question, answered by DeepSeek alone. The final list above combines all the AIs.

  1. 1
    Ivanti Connect Secure (2026)

    Multiple 2026 CVEs were exploited on the same day they became public — and the vendor’s median KEV lag is 0.8 days.

    Source ↗
  2. 2
    SonicWall SonicOS SSLVPN (2026)

    CVE-2026-19868 went from disclosure to confirmed exploitation in 0 days, against a 124-day median for the product.

  3. 3
    Microsoft SharePoint Server (2026)

    Microsoft’s overall median KEV lag is negative (−0.7 days), and SharePoint flaws have been exploited before or within days of patch Tuesday.

    Source ↗
  4. 4
    Cisco BroadWorks (2026)

    Cisco patched CVE-2026-20320 in August 2026; by November it was on CISA KEV, with a median Cisco KEV lag of 2 days.

    Source ↗
  5. 5
    WordPress Core (2026)

    CVE-2026-119977 was exploited in 3 days, faster than most enterprise products’ entire patch cycle.

  6. 6
    Atlassian Data Center (Confluence) (2026)

    CVE-2026-21589 reached CISA KEV on the day of publication, with a public PoC already referenced.

  7. 7
    GitLab Community/Enterprise Edition (2026)

    CVE-2026-97229 was exploited on the day it was published, with a PoC available within hours.

  8. 8
    PHP (2026)

    Two PHP2Shell flaws were exploited on the day of publication, with PoC references already circulating.

  9. 9
    Google Brotli (2026)

    A decoder flaw in a widely used compression library was exploited on publication day, showing supply-chain urgency.

  10. 10
    BeyondTrust Remote Support (2026)

    CVE-2026-102764 was added to KEV on 16 October 2026, the day after disclosure, with a PoC already referenced.

What DeepSeek noticed

The global median disclosure-to-exploitation window is now −7 days, meaning many products are attacked before a patch is even published.

Microsoft has the largest raw KEV count (390), but its median KEV lag is negative; the volume and the speed compound the risk.

Even niche or infrastructure products like LiteLLM have been exploited within 36 hours of advisory publication, showing the collapse is not limited to giant vendors.

🇨🇳 Wenxin10 places · 1 notes

The same question, answered by Wenxin alone. The final list above combines all the AIs.

  1. 1
    OpenSSH 2026

    Exploitation confirmed 0.5 days after public patch disclosure, fastest recorded in 24 months

  2. 2
    Google Chromium 2026

    Zero-day exploited publicly within 4 days of being reported to the vendor, 1 day after official patch release

  3. 3
    MikroTik RouterOS 7.x

    Added to CISA KEV 1 day after public exploit chain disclosure for the SSH path injection flaw

    Source ↗
  4. 4
    F5 BIG-IP APM 2026

    Added to CISA KEV the same day public unauthenticated RCE PoC was released

    Source ↗
  5. 5
    Check Point Security Management Server 2026

    Dual RCE flaws added to KEV the same day they were publicly disclosed with full PoC code

    Source ↗
  6. 6
    WordPress Core 4.7.0–7.1.1 2026

    LFI/RCE flaw added to KEV 3 days after the official patch was published

    Source ↗
  7. 7
    Microsoft Office 2024

    6 high-severity flaws published on the same day, 18.8% of all documented CVEs have public exploit code available

    Source ↗
  8. 8
    Microsoft SharePoint Server 2026

    Code injection RCE flaw added to KEV less than 7 days after public patch release

    Source ↗
  9. 9
    Cisco ISE 2026

    API gateway authentication bypass RCE exploited in the wild 2 days after official advisory publication

    Source ↗
  10. 10
    Apache Tomcat 2026

    HTTP/2 request smuggling PoC released publicly 3 days after patch for CVE-2026-86350

    Source ↗

What Wenxin noticed

Google security research confirmed in November 2026 that unassisted AI agents can now fully convert public software patches into working exploits with zero human input