Skip to navigation

Threat reportThreat IntelligenceTL-2026-0914

AI-Accelerated Exploitation Collapses Vulnerability-Management Patch Windows (Picus: ~24h time-to-exploit vs 43-day median fix)

highACTIVE

Picus Security's 2026 research documents an AI-driven structural shift in vulnerability management: AI-assisted adversaries have compressed median time-to-exploit (TTE) to roughly 24 hours (some sources cite ~10 hours) while the median organizational fix time for known-exploited vulnerabilities has risen to about 43 days.

Key facts

IdentifiersTL-2026-0914
Also known asAI Broke Vulnerability Management, 73 Seconds to Breach, Exploitation-Window Gap
SeverityHIGH
Affected productsIndustry-wide Vulnerability management and patch programs; Fortinet FortiGate (referenced as a mass-exploitation target in AWS Feb 2026 reporting); Mozilla Firefox (referenced as Anthropic Mythos exploit target)
ToolsAgentic Breach and Attack Simulation (BAS), Anthropic Claude 'Mythos' preview model, Autonomous pentesting
MITRE ATT&CK techniquesT1003 OS Credential Dumping, T1005 Data from Local System, T1046 Network Service Discovery, T1059 Command and Scripting Interpreter, T1068 Exploitation for Privilege Escalation, T1078 Valid Accounts, T1133 External Remote Services, T1190 Exploit Public-Facing Application, T1210 Exploitation of Remote Services, T1505 Server Software Component, T1556 Modify Authentication Process, T1587 Develop Capabilities and 4 more in the attack flow
MotivationUNKNOWN
Target sectorsfinancial, healthcare, government, technology, critical infrastructure, manufacturing
Target regionsGlobal, North America, Europe
Detection rules9 (1 shown free)
Sources8 cited sources

Detection coverage

As of 2026-06-23, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-0914 across Splunk SPL, Microsoft KQL and Sigma, covering 19 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: Credential Dumping + Run-Key Persistence Inside Exploitation Window — Post-Exploitation

Detects the second-45 credential-dump step (LSASS access / shadow read) and reversible HKCU Run-key persistence of the 73-second chain (T1003, T1547.001, T1068). Surfaces LSASS-targeting command lines and Run-key writes appearing on a host shortly after a public-facing-app exploit, independent of patch state.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1003, T1547.001, T1068
  • Data sources: sysmon, DeviceProcessEvents, DeviceRegistryEvents
  • Severity: critical
  • Confidence: high

Splunk SPL

index=sysmon sourcetype="XmlWinEventLog:Microsoft-Windows-Sysmon/Operational" (EventCode=1 OR EventCode=10 OR EventCode=13)
| eval cl=lower(CommandLine), tgt=lower(TargetObject), gt=lower(TargetImage)
| eval cred_dump=if(match(cl,"(lsass|sekurlsa|comsvcs\.dll.*minidump|procdump.*lsass)") OR like(gt,"%lsass.exe%"),1,0)
| eval shadow_read=if(match(cl,"(/etc/shadow|cat .*shadow|unshadow)"),1,0)
| eval runkey=if(like(tgt,"%currentversion\\run\\%"),1,0)
| where cred_dump=1 OR shadow_read=1 OR runkey=1
| eval signal=case(cred_dump=1,"credential_dump",shadow_read=1,"shadow_read",runkey=1,"runkey_persistence")
| stats values(signal) as signals, dc(signal) as signal_types, values(CommandLine) as cmds, values(TargetObject) as reg, min(_time) as first_seen, max(_time) as last_seen by Computer, User
| where signal_types>=1
| table first_seen, last_seen, Computer, User, signals, signal_types, cmds, reg

Microsoft KQL

let lookback = 24h;
union
( DeviceProcessEvents
  | where Timestamp > ago(lookback)
  | where ProcessCommandLine matches regex @"(?i)(lsass|sekurlsa|comsvcs\.dll.*minidump|procdump.*lsass|/etc/shadow|unshadow)"
  | extend Signal = "credential_dump", Detail = ProcessCommandLine ),
( DeviceRegistryEvents
  | where Timestamp > ago(lookback)
  | where ActionType == "RegistryValueSet"
  | where RegistryKey has @"\CurrentVersion\Run"
  | extend Signal = "runkey_persistence", Detail = strcat(RegistryKey, "\\", RegistryValueName) )
| summarize Signals = make_set(Signal, 5), Details = make_set(Detail, 20), FirstSeen = min(Timestamp), LastSeen = max(Timestamp) by DeviceName
| project FirstSeen, LastSeen, DeviceName, Signals, Details
| order by FirstSeen asc

Sigma

title: Credential Dumping and Run-Key Persistence in Compressed Exploitation Window
id: 4a1f9c2e-0b21-4d7a-9e3c-5f1a2b3c4d03
status: experimental
description: Detects LSASS-targeting credential-access command lines, /etc/shadow reads, or HKCU Run-key persistence consistent with the second-45 step of the TL-2026-0914 73-second AI breach chain.
references:
    - https://www.picussecurity.com/resource/report/ai-broke-vulnerability-management
author: Threadlinqs Team
date: 2026/06/23
tags:
    - attack.credential_access
    - attack.t1003
    - attack.persistence
    - attack.t1547.001
    - attack.privilege_escalation
    - attack.t1068
logsource:
    product: windows
    category: process_creation
detection:
    selection_creddump:
        CommandLine|contains:
            - 'lsass'
            - 'sekurlsa'
            - 'comsvcs.dll'
            - 'procdump'
            - '/etc/shadow'
            - 'unshadow'
    condition: selection_creddump
falsepositives:
    - Legitimate memory-diagnostic or backup tooling accessing LSASS
    - Authorized red-team / BAS credential-access validation
level: critical

Known false positives:

  • Legitimate memory-diagnostic or backup tooling accessing LSASS
  • Authorized red-team / BAS credential-access validation runs

Tuning notes: Allowlist sanctioned EDR/backup processes that legitimately read LSASS. Raise fidelity by alerting only when a credential-dump signal co-occurs on a host with an SPL-001/SPL-002 event within the same hour.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • Edge Appliance Exploit Burst Inside Patch SLA — Exploitation · unlock with TL Creds
  • Compressed Kill Chain on Single Host Within One Hour — Post-Exploitation · unlock with TL Creds
  • MFA Bypass / Session-Token Replay in Automated Chain — Credential Access · unlock with TL Creds
  • AI Parallel Mass-Exploitation Fan-Out From Single Source — C2 / Lateral Movement · unlock with TL Creds
  • Run-Key Persistence Created During Active Exploitation — Post-Exploitation · unlock with TL Creds
  • Web-Server Process Spawning Command Interpreter — C2 / Lateral Movement · unlock with TL Creds
  • AI-Accelerated Public-Facing App Exploitation on Edge/VPN Appliance — Exploitation · unlock with TL Creds
  • Web Shell Write in Web Root Following Edge Exploitation — Post-Exploitation · unlock with TL Creds

Indicators of compromise

As of 2026-06-23, Threadlinqs Intelligence tracks 19 indicators of compromise for TL-2026-0914.

19 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

The resulting exploitation-window gap renders patch-velocity spending diminishing in value and motivates a shift of budget toward continuous control validation via Breach and Attack Simulation (BAS) and autonomous validation.

This is a TREND / threat-landscape item rather than a single-CVE threat. Picus Security (Picus Labs), in 'AI Broke Vulnerability Management: The CISO's Case for Moving Budget to BAS' and the companion BleepingComputer analysis '73 Seconds to Breach, 24 Hours to Patch,' both authored by Picus Security Research Engineer Sila Ozeren Hacioglu, argue that generative and agentic AI have fundamentally inverted the attacker/defender timing economics of vulnerability management.

The core quantitative claim is a widening exploitation-window gap. Median time-to-exploit (TTE) has collapsed from roughly 2.3 years in 2018, to ~56 days in 2024, ~23 days in 2025, and approximately 24 hours (with some measurements near ~10 hours median) in 2026, as tracked by the Zero Day Clock. Over the same period the median fix time for known-exploited vulnerabilities rose from ~32 to ~43 days, the share of vulnerabilities fully patched fell from 38% to 26%, and the median number of known-exploited vulnerabilities carried per organization rose from 11 (2024) to 16 (2025). Even top-performing organizations close only 30-40% of known-exploited vulnerabilities within the first week. The Verizon 2026 DBIR attributes roughly 32% of breach initial-access techniques to vulnerability exploitation.

The reports anchor the trend in concrete AI-exploitation evidence. Anthropic's Claude 'Mythos' preview model (April 2026) generated 181 working Firefox exploits in 14 days versus a prior state-of-the-art of 2, surfaced thousands of zero-days across major operating systems and browsers, and identified a 27-year-old OpenBSD bug; over 99% of what Mythos found remained unpatched at publication. In one month a Mythos-class capability reportedly found 10,000+ high/critical vulnerabilities. Separately, an AWS February 2026 threat-intelligence report described a single AI-augmented operator impacting 2,516 FortiGate devices across 106 countries (an earlier figure cited 600+ FortiGate devices across 55+ countries), executing attacks in parallel in minutes per target.

The central illustrative scenario is a 73-second AI-driven breach: at second 5 a CVE is exploited, at second 20 MFA is bypassed, at second 30 a web shell is deployed, at second 45 credentials are dumped, and at second 73 the compromise is complete. The defensive response chain by contrast unfolds over a SIEM alert (minute 1), Tier-1 pickup (minute 5), SOAR playbook (minute 15), a Jira ticket (hour 1), IT-ops queue (hour 4+), and patch deployment only at hour 24 - the next day.

Picus' prescriptive thesis is to reallocate budget from chasing patch velocity (diminishing returns at scale) toward continuous validation of control effectiveness using agentic Breach and Attack Simulation and autonomous pentesting, framed within CTEM (Continuous Threat Exposure Management) and the NIST Cybersecurity Framework (Identify / Protect / Validate). Notably, Picus' agentic BAS is described as matching fresh threat reports against a curated, pre-vetted library of safe, ready-made test building blocks rather than asking AI to author live payloads. Vendor-stated outcomes for this approach include 2X control effectiveness within three months and 89% lower MTTR; these are vendor claims, not independently verified metrics. Picus was named an Innovation Leader for Automated Security Validation in Frost & Sullivan's Frost Radar 2026. No specific CVE, IOC, malware family, or single threat actor is named as the subject of this item by design - it is a landscape/trend record.

MITRE ATT&CK techniques by tactic:

Credential Access

T1003 OS Credential Dumping; T1621 Multi-Factor Authentication Request Generation

Collection

T1005 Data from Local System

Discovery

T1046 Network Service Discovery

Execution

T1059 Command and Scripting Interpreter

Privilege Escalation

T1068 Exploitation for Privilege Escalation

Initial Access

T1078 Valid Accounts; T1133 External Remote Services; T1190 Exploit Public-Facing Application

Lateral Movement

T1210 Exploitation of Remote Services

Persistence

T1505 Server Software Component

defense-impairment

T1556 Modify Authentication Process

Resource Development

T1587 Develop Capabilities; T1588 Obtain Capabilities

Reconnaissance

T1595 Active Scanning; T1596 Search Open Technical Databases

Remediation

Patches

  • Maintain an exploit-prioritized patch pipeline targeting sub-week remediation for known-exploited and internet-facing vulnerabilities
  • Patch perimeter/edge devices (e.g., FortiGate-class VPN/firewall appliances) on an emergency cadence given AI-accelerated mass exploitation

Immediate actions

  • Prioritize remediation of CISA KEV-listed and actively-exploited vulnerabilities within the first 24-72 hours, not on a multi-week SLA
  • Treat exploitability and active-exploitation signals (KEV, VulnCheck, Zero Day Clock) above raw CVSS when prioritizing
  • Deploy virtual patching / WAF and IPS signatures as a stopgap for internet-facing assets while permanent patches are scheduled

Workarounds

  • Enforce phishing-resistant MFA and conditional access to raise the cost of MFA-bypass steps in automated chains
  • Monitor for and block web shell deployment and credential-dumping behaviors with EDR behavioral detection independent of patch state

Longer-term hardening

  • Adopt continuous control validation via agentic Breach and Attack Simulation (BAS) to measure whether deployed controls actually block and detect current threats
  • Operationalize CTEM (Continuous Threat Exposure Management) to continuously scope, discover, prioritize, validate, and mobilize
  • Add autonomous/automated pentesting to validate exploit chains end-to-end rather than relying on point-in-time assessments
  • Reduce internet-facing attack surface and enforce network segmentation to limit blast radius of fast exploitation

Affected and fixed versions

  • Industry-wide — Vulnerability management and patch programs
    Vulnerable versions: organizations relying on patch-velocity SLAs
    Fixed in: organizations running continuous control validation (BAS/CTEM)
  • Fortinet — FortiGate (referenced as a mass-exploitation target in AWS Feb 2026 reporting)
    Vulnerable versions: internet-facing FortiGate appliances
    Fixed in: patched per vendor advisory
  • Mozilla — Firefox (referenced as Anthropic Mythos exploit target)
    Vulnerable versions: versions targeted by AI-generated exploits
    Fixed in: current patched releases

Timeline

  • Median CVE-to-exploit (time-to-exploit) measured at roughly 2.3 years, per Zero Day Clock / Picus historical baseline.
  • Median time-to-exploit falls to ~53-56 days; median known-exploited vulnerabilities carried per organization is 11; 38% of vulnerabilities fully patched.
  • Median time-to-exploit drops to ~23 days; median known-exploited vulnerabilities per organization rises to 16.
  • AWS February 2026 threat-intelligence reporting describes a single AI-augmented operator impacting 2,516 FortiGate devices across 106 countries (earlier figure: 600+ devices across 55+ countries), executing attacks in parallel within minutes per target.
  • Anthropic's Claude 'Mythos' preview model generates 181 working Firefox exploits in 14 days (prior state-of-the-art: 2), surfaces thousands of zero-days across OSes/browsers, and finds a 27-year-old OpenBSD bug; over 99% remain unpatched at publication.
  • Picus Autonomous Validation Summit held (May 12 and 14, 2026) presenting the autonomous-validation thesis.
  • BleepingComputer publishes '73 Seconds to Breach, 24 Hours to Patch: The Case for Autonomous Validation' by Picus' Sila Ozeren Hacioglu, detailing the 73-second AI breach scenario and the ~24h-to-patch response chain.
  • Picus Security publishes 'AI Broke Vulnerability Management: The CISO's Case for Moving Budget to BAS'; covered by The Hacker News the same day. Median TTE ~24h, median fix 43 days, 26% fully patched, Verizon 2026 DBIR attributing 32% of initial access to vulnerability exploitation.
  • Threadlinqs Intelligence creates landscape/trend record TL-2026-0914 documenting the AI-accelerated exploitation-window gap and the BAS/CTEM defensive response.

Sources cited

Previous and next in threat intel

Drafted by the Threadlinqs research pipeline from 8 sources; published ; last updated —.

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats