Threat reportThreat IntelligenceTL-2026-1214
Rapid7 Policy Paper 'Modernizing Global Vulnerability Standards' Warns AI-Driven Vulnerability Discovery Is Outpacing CVE/CVSS/NVD Standards
Rapid7 published a policy paper urging reform of CVE, CVSS, NVD, CISA KEV, and EPSS after presenting findings to a White House private-sector consultation in June 2026, arguing that production AI systems from Anthropic, OpenAI, and Google DeepMind can now discover, chain, and remediate vulnerabilities at machine speed that legacy vulnerability-management infrastructure cannot absorb.
Key facts
| Identifiers | TL-2026-1214 |
|---|---|
| Also known as | Modernizing Global Vulnerability Standards |
| Severity | INFORMATIONAL |
| Affected products | NIST National Vulnerability Database (NVD) |
| Tools | Anthropic Mythos Preview, OpenAI GPT-5.5-Cyber |
| MITRE ATT&CK techniques | T1005 Data from Local System, T1027 Obfuscated Files or Information, T1041 Exfiltration Over C2 Channel, T1046 Network Service Discovery, T1068 Exploitation for Privilege Escalation, T1071 Application Layer Protocol, T1078 Valid Accounts, T1190 Exploit Public-Facing Application, T1195 Supply Chain Compromise, T1203 Exploitation for Client Execution, T1210 Exploitation of Remote Services, T1211 Exploitation for Stealth and 9 more in the attack flow |
| Motivation | UNKNOWN |
| Target sectors | government administration, technology, critical-infrastructure, open-source-ecosystem |
| Target regions | united states of america, Global |
| Detection rules | 9 (1 shown free) |
| Sources | 20 cited sources |
Detection coverage
As of 2026-07-11, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-1214 across Splunk SPL, Microsoft KQL and Sigma, covering 21 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: Automated Vuln-Feed Recon from Scripting Host — Exploitation Prep
Detects a scripting interpreter (powershell.exe, python.exe, bash) issuing outbound HTTP(S) requests to hostnames containing vuln-feed/cve-feed/nvd/kev-style patterns, consistent with AI-paced automated recon that scrapes public vulnerability databases before targeting a public-facing application (T1596, T1592, T1190). Excludes known vulnerability-scanner processes.
- Languages: Splunk SPL, Microsoft KQL, Sigma
- ATT&CK: T1596, T1592, T1190
- Data sources: sysmon, windows_security, deviceNetworkEvents
- Severity: medium
- Confidence: medium
Splunk SPL
index=sysmon sourcetype="XmlWinEventLog:Microsoft-Windows-Sysmon/Operational" EventCode=3
| eval proc=lower(Image)
| where (like(proc, "%powershell.exe") OR like(proc, "%pwsh.exe") OR like(proc, "%python.exe") OR like(proc, "%bash%"))
| where match(DestinationHostname, "(?i)(vuln-feed|cve-feed|nvd\\.nist|cve\\.org|kev-feed)")
| where NOT (like(proc, "%nessus.exe") OR like(proc, "%qualys.exe") OR like(proc, "%rapid7%") OR like(proc, "%nexpose%"))
| stats count min(_time) as first_seen max(_time) as last_seen by host, proc, DestinationHostname, User
| where count > 0Microsoft KQL
DeviceNetworkEvents
| where Timestamp > ago(24h)
| where InitiatingProcessFileName in~ ("powershell.exe", "pwsh.exe", "python.exe", "bash")
| where RemoteUrl matches regex @"(?i)(vuln-feed|cve-feed|nvd\.nist|cve\.org|kev-feed)"
| where InitiatingProcessFileName !in~ ("nessus.exe", "qualys.exe", "nexpose.exe")
| project Timestamp, DeviceName, InitiatingProcessFileName, InitiatingProcessAccountName, RemoteUrl, RemoteIP
| summarize count(), min(Timestamp), max(Timestamp) by DeviceName, InitiatingProcessFileName, RemoteUrlSigma
title: Suspicious Outbound Requests to Vulnerability/CVE Feed Endpoints from Non-Scanner Process
id: 7a3d9e10-1214-4a01-8b1e-2f0e9c1f0001
status: experimental
description: Detects a scripting interpreter reaching out to vulnerability/CVE-feed-style hostnames, consistent with AI-paced automated vulnerability-discovery recon preceding exploitation of a public-facing application.
references:
- https://www.rapid7.com/blog/post/ai-modernizing-global-vulnerability-standards
author: Threadlinqs Team
date: 2026-07-11
tags:
- attack.reconnaissance
- attack.t1596
- attack.t1592
- attack.initial-access
- attack.t1190
logsource:
category: network_connection
product: windows
detection:
selection:
DestinationHostname|contains:
- 'vuln-feed'
- 'cve-feed'
- 'nvd.nist'
- 'kev-feed'
Image|endswith:
- '\powershell.exe'
- '\pwsh.exe'
- '\python.exe'
filter:
Image|endswith:
- '\nessus.exe'
- '\qualys.exe'
- '\nexpose.exe'
condition: selection and not filter
falsepositives:
- Authorized vulnerability-management tooling built on scripting runtimes (allowlist known automation accounts/hosts)
- Internal threat-intel harnesses (such as this platform) that legitimately poll CVE feeds via scripts
level: mediumKnown false positives:
- Authorized internal threat-intel or vuln-management automation (e.g. this harness's own HUNT phase) that legitimately scripts CVE-feed polling
Tuning notes: Maintain an allowlist of known internal vulnerability-management/automation hosts and service accounts; tighten the hostname regex to your org's specific feed allow/deny list to reduce noise from legitimate CVE-ingestion pipelines.
- Bulk Installed-Software Enumeration Preceding External Network Activity · unlock with TL Creds
- Unauthorized Valid-Account Use Following Chaining-Risk Artifact Creation · unlock with TL Creds
- Obfuscated Enrichment/Exfil Script Execution Following Recon-to-Discovery Chain · unlock with TL Creds
- Beaconing C2 Callback to Vuln-Feed-Styled Domain (Chaining/Exfil Channel) · unlock with TL Creds
- Lateral Movement via Exploitation of Remote Service Following Chained Finding · unlock with TL Creds
- Server Software Component Persistence via Rapid Post-Discovery Deployment · unlock with TL Creds
- Supply-Chain Package Compromise Attempt Following Automated Recon · unlock with TL Creds
- Client-Side Exploitation Following Fingerprinted Browser/OS Recon · unlock with TL Creds
Indicators of compromise
As of 2026-07-11, Threadlinqs Intelligence tracks 21 indicators of compromise for TL-2026-1214.
Technical analysis
On June 29, 2026, Rapid7 published 'Modernizing Global Vulnerability Standards,' authored by Corey Thomas (Chairman & CEO) and Sabeen Malik (VP, Global Government Affairs & Public Policy), and presented its findings to the White House during a private-sector consultation earlier in June 2026. The paper's central claim is that the legacy vulnerability-disclosure and scoring ecosystem — CVE identifiers, CVSS scoring, the National Vulnerability Database (NVD), the CISA Known Exploited Vulnerabilities (KEV) catalog, the Exploit Prediction Scoring System (EPSS), and the Vulnerabilities Equities Process — was designed for human-paced research and can no longer absorb AI-paced vulnerability discovery.
The paper cites Stanford HAI's AI Index 2026 finding that unguided AI agent solve rates on the Cybench cybersecurity benchmark rose from 15% in 2024 to 93% in 2025, and that CVE submissions grew 263% between 2020 and 2025 from human-driven discovery alone — before accounting for AI-assisted discovery. It notes that in April 2026, Anthropic, OpenAI, and Google DeepMind each announced production-grade AI systems capable of discovering, chaining, and in some cases remediating software vulnerabilities: Anthropic's 'Mythos Preview' model found high-severity vulnerabilities in every major operating system and web browser; OpenAI expanded its 'Daybreak' cybersecurity initiative with a full release of GPT-5.5-Cyber and launched 'Patch the Planet' with Trail of Bits to help open-source maintainers triage and repair vulnerabilities; and the Linux Foundation received $12.5M in grants from Anthropic, AWS, GitHub, Google, Google DeepMind, Microsoft, and OpenAI (managed via Alpha-Omega and OpenSSF) to help maintainers absorb the resulting influx of findings.
The paper's urgency is reinforced by two independent government-side signals cited in the surrounding coverage: (1) on April 15, 2026, NIST announced it could no longer keep pace with 'record growth' in CVE submissions and shifted NVD to a risk-based triage model — prioritizing enrichment only for CVEs in the CISA KEV catalog, CVEs affecting software used within the federal government, and CVEs for 'critical software' as defined by Executive Order 14028, while moving all backlogged pre-March-2026 CVEs to a 'Not Scheduled' category; and (2) a May 26, 2026 Department of Commerce Office of Inspector General report (OIG-26-020-I) found NVD severity scores matched independent evaluators only 12% of the time (i.e., wrong roughly 88% of the time), that the unprocessed backlog grew from ~13,000 entries in June 2024 to over 27,000 by end of 2025, and that in the 30 days before April 7, 2026 the NVD served roughly 300,000 unique users downloading an average of 22 terabytes of data per day — a scale mismatch the IG called a threat to 'the NVD's utility and public trust.' NIST was ordered to submit a corrective action plan addressing all six IG recommendations by July 31, 2026.
The paper's timing also follows a June 23, 2026 Five Eyes cyber security agencies joint statement warning that AI models capable of launching major cyberattacks able to overwhelm government and business defenses are 'months, not years' away, and that AI is shortening the time between vulnerability discovery and exploitation — a framing Rapid7 leverages to argue reform cannot wait for the next multi-year standards cycle.
Rapid7's recommended reforms: (1) formally recognize verified AI-demonstrated exploitability as a distinct evidentiary category in CVE/NVD records; (2) add chaining-risk metadata to vulnerability records so defenders can see when a low/medium-severity finding is part of a viable multi-step exploit chain; (3) require reachability guidance accompanying AI-discovered findings (i.e., whether the vulnerable code path is actually reachable/exploitable in a given deployment, not merely present); (4) update the Vulnerabilities Equities Process to account for AI-discovered zero-days; (5) direct new investment into CVE and NVD infrastructure; (6) establish standardized capability disclosure requirements from frontier AI labs regarding vulnerability-discovery capabilities; (7) strengthen international coordination with CISA in a leadership role; and (8) implement three access/verification standards for AI-discovered vulnerability data — independent verification, broad curated access, and rigorous data-quality standards.
This is not a discrete exploited vulnerability — there is no associated CVE, CVSS score, or PoC — but an industry policy signal with direct downstream effects on this harness's own triage assumptions: CVSS/NVD completeness can no longer be treated as authoritative or timely, EPSS and KEV become comparatively more load-bearing for prioritization, and 'AI-demonstrated exploitability' and 'chaining-risk' are emerging as new first-class attributes that vulnerability records (including this platform's own) may need to carry going forward.
MITRE ATT&CK techniques by tactic:
Collection
Defense Evasion
T1027 Obfuscated Files or Information; T1211 Exploitation for Stealth
Exfiltration
T1041 Exfiltration Over C2 Channel
Discovery
T1046 Network Service Discovery; T1518 Software Discovery
Privilege Escalation
T1068 Exploitation for Privilege Escalation
Command and Control
T1071 Application Layer Protocol
Persistence
T1078 Valid Accounts; T1505 Server Software Component
Initial Access
T1190 Exploit Public-Facing Application; T1195 Supply Chain Compromise
Execution
T1203 Exploitation for Client Execution
Lateral Movement
T1210 Exploitation of Remote Services
Credential Access
Impact
Resource Development
T1585 Establish Accounts; T1587 Develop Capabilities; T1588 Obtain Capabilities
Reconnaissance
T1592 Gather Victim Host Information; T1596 Search Open Technical Databases
Remediation
Immediate actions
- Do not treat NVD CVSS enrichment completeness as authoritative for CVEs published or backlogged before March 1, 2026 — NIST has moved those to 'Not Scheduled' status
- Cross-check NVD severity scores against independent scoring sources (vendor advisories, VulnCheck, CVSS calculated by internal analysts) given the DOC OIG finding that NVD scores matched independent evaluators only 12% of the time
- Weight CISA KEV membership and EPSS score more heavily than raw CVSS/NVD enrichment status when triaging, since NIST now only guarantees enrichment for KEV-listed, federally-used, or EO-14028 'critical software' CVEs
- Monitor Rapid7, NIST, and CISA channels for the July 31, 2026 NIST corrective action plan deadline responding to DOC OIG report OIG-26-020-I
Longer-term hardening
- Track emergence of 'AI-demonstrated exploitability' and 'chaining-risk metadata' as new vulnerability-record attributes and evaluate incorporating equivalent fields into this platform's threat schema
- Build internal reachability-analysis capability for ingested CVEs rather than depending solely on external reachability guidance, given the pace mismatch Rapid7 and NIST both describe
- Establish a recurring watch on frontier AI lab capability disclosures (Anthropic, OpenAI, Google DeepMind) for vulnerability-discovery/remediation systems, since these are becoming a primary source of net-new findings outpacing human-driven CVE submission
Affected and fixed versions
- NIST — National Vulnerability Database (NVD)
Vulnerable versions: pre-2026-04-15 enrichment/triage model
Fixed in: risk-based triage model effective 2026-04-15; corrective action plan due 2026-07-31 per DOC OIG-26-020-I
Timeline
- Atlantic Council publishes commentary warning new AI models are pushing open-source security triage capacity to its limits, ahead of the coordinated April AI-lab announcements.
- Anthropic's 'Mythos Preview' model is reported to have found high-severity vulnerabilities in every major operating system and web browser, part of a wave of April 2026 frontier-lab vulnerability-discovery announcements.
- Stanford HAI publishes the AI Index 2026 report, documenting that unguided AI agent solve rates on the Cybench cybersecurity benchmark rose from 15% (2024) to 93% (2025).
- NIST announces NVD can no longer keep pace with record CVE growth and moves to a risk-based triage model, prioritizing only KEV-listed, federally-used, and EO-14028 critical-software CVEs; pre-March-2026 backlog moved to 'Not Scheduled.'
- Department of Commerce Office of Inspector General publishes report OIG-26-020-I finding NVD severity scores matched independent evaluators only 12% of the time and the unprocessed backlog grew from ~13,000 (June 2024) to over 27,000 (end 2025).
- Trade press (The Record, Help Net Security, CyberScoop, CSO Online, BankInfoSecurity) widely covers the DOC OIG findings on NIST/NVD mismanagement.
- Five Eyes cyber security agencies issue a joint statement warning that AI models capable of overwhelming government/business cyber defenses are 'months, not years' away, and that AI is shortening the discovery-to-exploitation window.
- Rapid7 publishes 'Modernizing Global Vulnerability Standards' by Corey Thomas and Sabeen Malik, having presented the findings to a White House private-sector consultation earlier in June 2026, recommending reforms to CVE, CVSS, NVD, KEV, EPSS, and the Vulnerabilities Equities Process.
- Deadline for NIST to submit a corrective action plan to the DOC Inspector General addressing all six recommendations from report OIG-26-020-I.
Sources cited
Showing 15 of 20 sources, primary sources first.
- NIST Updates NVD Operations to Address Record CVE Growth NIST · Advisory · 2026-04-15
- Five Eyes Cyber Security Agencies Statement (CISA) CISA · Government advisory · 2026-06-23
- Five Eyes Cyber Security Agencies Statement (NSA) nsa.gov · Advisory · 2026-06-23
- U.S. Department of Commerce Office of Inspector General Report OIG-26-020-I oig.doc.gov · Research · 2026-05-26
- Inside the AI Index: 12 Takeaways from the 2026 Report hai.stanford.edu · Research · 2026-04-14
- The 2026 AI Index Report hai.stanford.edu · Research · 2026-04-14
- New AI models are pushing open-source security to its limits. Their developers must step up. atlanticcouncil.org · Research · 2026-04-01
- AI Is Modernizing Global Vulnerability Standards Rapid7 · Blog · 2026-06-29
- NIST Can't Keep Up. The Whole Digital Ecosystem Will Soon Feel It. justsecurity.org · Blog · 2026-04-20
- Five Eyes Intelligence Alliance Warns AI-Fueled Cyberattacks Are Just Months Away kiteworks.com · Blog · 2026-06-23
- NIST Drops NVD Enrichment for Pre-March 2026 Vulnerabilities infosecurity-magazine.com · News · 2026-04-16
- NIST admits defeat on NVD backlog, will enrich only highest-risk CVEs going forward Help Net Security · News · 2026-04-16
- NIST National Vulnerability Database Severity Scores Wrong 88% of Time, Inspector General Finds techtimes.com · News · 2026-06-02
- Inspector general finds NIST mistakes have made vulnerability database ineffective The Record · News · 2026-06-01
- How NIST fumbled management of the National Vulnerability Database Help Net Security · News · 2026-06-01
Previous and next in threat intel
Drafted by the Threadlinqs research pipeline from 20 sources; published ; last updated —.