Skip to navigation

Threat reportThreat IntelligenceTL-2026-0909

OpenAI Releases GPT-5.5-Cyber: Defender-Restricted AI for Automated Vulnerability Detection, Exploitability Validation, and Patching (Daybreak / Patch the Planet)

MONITORING

OpenAI released GPT-5.5-Cyber, a cyber-specialized model distributed only to verified, trusted defenders through its Trusted Access for Cyber program, alongside the Codex Security plugin and the Patch the Planet initiative co-built with Trail of Bits and HackerOne. The model can navigate large codebases, trace attack paths, validate exploitability, and generate targeted patches in a single automated workflow.

Key facts

IdentifiersTL-2026-0909
Also known asGPT-5.5-Cyber, Daybreak, Patch the Planet, Codex Security
SeverityINFO
Affected productsOpenAI GPT-5.5-Cyber (Daybreak / Trusted Access for Cyber)
ToolsCodex Security, Daybreak, GPT-5.5-Cyber
MITRE ATT&CK techniquesT1190 Exploit Public-Facing Application, T1195.001 Compromise Software Dependencies and Development Tools, T1203 Exploitation for Client Execution, T1587 Develop Capabilities, T1587.004 Exploits, T1588 Obtain Capabilities, T1588.002 Tool, T1588.005 Exploits, T1588.006 Vulnerabilities, T1592 Gather Victim Host Information, T1592.002 Software, T1593 Search Open Websites/Domains and 4 more in the attack flow
MotivationUNKNOWN
Target sectorstechnology, open-source software, government, critical infrastructure
Target regionsGlobal, North America, Europe, Asia-Pacific
Detection rules9 (1 shown free)
Sources10 cited sources

Detection coverage

As of 2026-06-23, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-0909 across Splunk SPL, Microsoft KQL and Sigma, covering 15 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: Agentic AI Vuln-Research — High-Fanout Source-Read Burst (Exploitation/Recon)

Detects a single process performing a high-fanout recursive read sweep over hundreds of source files (.c/.py/.go/.java) in a short window — the bulk codebase-ingestion footprint an agentic AI vulnerability-research workflow (GPT-5.5-Cyber / Codex Security class) leaves when it navigates a large codebase to trace attack paths. Maps to the Reconnaissance 'Gather Victim Host Information: Software' surface.

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1592.002, T1596.005
  • Data sources: sysmon, microsoft_defender_endpoint, linux_auditd
  • Severity: low
  • Confidence: medium

Splunk SPL

index=sysmon sourcetype="XmlWinEventLog:Microsoft-Windows-Sysmon/Operational" EventCode=11
| eval ext=lower(replace(TargetFilename, "^.*\.", ""))
| search ext IN ("c","h","cpp","py","go","java","rs","js","ts")
| bin _time span=10s
| stats dc(TargetFilename) AS files_touched values(TargetFilename) AS sample BY _time, ProcessId, Image
| where files_touched > 200
| sort - files_touched

Microsoft KQL

DeviceFileEvents
| where Timestamp > ago(24h)
| where ActionType in ("FileAccessed", "FileModified", "FileCreated")
| where FileName has_any (".c", ".h", ".cpp", ".py", ".go", ".java", ".rs", ".js", ".ts")
| summarize files_touched = dcount(FolderPath) by bin(Timestamp, 10s), InitiatingProcessId, InitiatingProcessFileName
| where files_touched > 200
| order by files_touched desc

Sigma

title: Agentic AI Vuln-Research Footprint - High-Fanout Source Read Burst
id: 5b1f8a40-09c1-4e7a-9a11-0909a0010001
status: experimental
description: Single process reads hundreds of source files in seconds, modeling bulk codebase ingestion by an AI vuln-research agent.
references:
    - https://openai.com/index/daybreak-securing-the-world/
    - https://attack.mitre.org/techniques/T1592/002/
author: Threadlinqs Team
date: 2026/06/23
tags:
    - attack.reconnaissance
    - attack.t1592.002
    - attack.resource_development
logsource:
    product: linux
    category: file_event
detection:
    selection:
        Image|endswith:
            - '/cat'
            - '/grep'
            - '/find'
            - '/python3'
        TargetFilename|endswith:
            - '.c'
            - '.py'
            - '.go'
            - '.java'
    timeframe: 10s
    condition: selection | count(TargetFilename) by Image > 200
falsepositives:
    - Legitimate IDE indexing, source-tree backup, or CI checkout/build steps
level: low

Known false positives:

  • IDE/editor source indexing (VS Code, JetBrains)
  • CI/CD checkout and build pipelines reading the full source tree
  • Backup or AV full-disk scans

Tuning notes: Tune files_touched threshold (200) to your repo size; baseline known IDE/CI process names and exclude them. Raise the time bin to 30s on noisy hosts.

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)
  • Interpreter Spawning Burst of Static-Analysis Children (Attack-Path Tracing) · unlock with TL Creds
  • Source Files Clustered by Dangerous-Sink Signature (Variant Analysis) · unlock with TL Creds
  • Unroutable / TEST-NET AI-Egress Connection Attempt (C2 Beacon Modeling) · unlock with TL Creds
  • AI Vuln-Research Lab Marker — Registry Persistence Value (Sigma) · unlock with TL Creds
  • Vulnerability-Scanning Recon Against Public-Facing Software (Sigma) · unlock with TL Creds
  • Software Supply-Chain Dependency Tampering Around AI-Patch Workflow (Sigma) · unlock with TL Creds
  • Exploit-Candidate / Remediation-Evidence Marker Artifact Written (Post-Exploitation) · unlock with TL Creds
  • Bulk Source Read Followed by AI-Inference HTTPS Egress (C2 / Exfil Channel) · unlock with TL Creds

Indicators of compromise

As of 2026-06-23, Threadlinqs Intelligence tracks 15 indicators of compromise for TL-2026-0909.

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

Technical analysis

This is a defensive/dual-use AI capability announcement — not an active exploit, malware campaign, or vulnerability.

On June 22-23, 2026, OpenAI expanded its Daybreak cybersecurity platform with the full release of GPT-5.5-Cyber, the Codex Security plugin, and the Patch the Planet initiative. GPT-5.5-Cyber is a cyber-specialized variant that, per OpenAI and reporting by Cyber Security News and Axios, can navigate large codebases, trace attack paths, validate exploitability, generate targeted patches, and produce remediation evidence within a single automated workflow. It reports the highest single-model score recorded on CyberGym at 85.6% (versus 81.8% for GPT-5.5), 39.5% on ExploitGym (versus 25.95% for GPT-5.5), and 69.8% on SEC-bench Pro (versus 63.1% for GPT-5.5).

The full model is not available for general use. It is limited to verified, trusted defenders through OpenAI's Trusted Access for Cyber program, which reduces automated safety refusals for approved defensive tasks such as secure code review, vulnerability triage, malware analysis, red teaming, and penetration testing. GPT-5.5 with Trusted Access is recommended as the entry point. OpenAI states it conducted pre-deployment testing with the Center for AI Standards and Innovation (CAISI), coordinated with the Office of the National Cyber Director (ONCD), and references the June 2026 Executive Order on AI security. Trusted Access for Cyber partnerships named include Australia, Canada, France, Germany, Japan, South Korea, EU institutions, and ENISA. Daybreak partner roster includes Akamai, Cisco, Cloudflare, CrowdStrike, Fortinet, Oracle, Palo Alto Networks, and Zscaler.

The Codex Security plugin, launched in research preview in March 2026, can review recent commits and produce reports with severity, affected code locations, validation evidence, and remediation guidance; trace attack paths; build threat models; validate findings; generate patches; and export results into vulnerability management systems. Reported aggregate Codex Security statistics include scanning over 30 million commits, processing more than 30,000 codebases, handling over 70,000 manually verified fixes, and automatically resolving over 500,000 findings. The Patch the Planet initiative, co-founded with Trail of Bits and partnered with HackerOne, pairs AI-assisted vulnerability research with human expert review; an initial five-day sprint surfaced hundreds of issues, merged dozens of patches, and produced reusable testing workflows including fuzzing, variant analysis, and differential testing. More than 30 open-source projects have committed to participate, with initial participants including cURL, Go, Python, Sigstore, and pyca/cryptography.

From a threat-intelligence standpoint this is tracked as an INFORMATIONAL industry/landscape signal documenting a dual-use AI capability. The same capabilities that accelerate defensive vulnerability discovery, exploitability validation, and patch generation also lower the barrier to offensive vulnerability research and exploit development if misused — which is why OpenAI gates the full model behind verified-defender access and reduced-refusal Trusted Access controls. There is no CVE, no active exploitation, no public proof-of-concept against a target, no malware family, no threat actor, and no malicious indicators of compromise associated with this record. The MITRE techniques and indicators below document the dual-use capability surface (primarily Resource Development and Reconnaissance) and the benign reference entities, not an observed attack.

MITRE ATT&CK techniques by tactic:

Initial Access

T1190 Exploit Public-Facing Application; T1195.001 Compromise Software Dependencies and Development Tools

Execution

T1203 Exploitation for Client Execution

Resource Development

T1587 Develop Capabilities; T1587.004 Exploits; T1588 Obtain Capabilities; T1588.002 Tool; T1588.005 Exploits; T1588.006 Vulnerabilities

Reconnaissance

T1592 Gather Victim Host Information; T1592.002 Software; T1593 Search Open Websites/Domains; T1595 Active Scanning; T1595.002 Vulnerability Scanning; T1596 Search Open Technical Databases; T1596.005 Scan Databases

Remediation

Immediate actions

  • No remediation required: this is a defensive/dual-use AI capability announcement, not a vulnerability or active threat.
  • Defenders eligible for OpenAI Trusted Access for Cyber can evaluate GPT-5.5-Cyber and Codex Security for secure code review and vulnerability triage.

Longer-term hardening

  • Incorporate AI-assisted vulnerability discovery (fuzzing, variant analysis, differential testing) into secure SDLC and patch management workflows.
  • Track dual-use AI capability uplift as a landscape factor in threat models: assume adversaries gain similar AI-accelerated vulnerability research over time and prioritize timely patching of public-facing and open-source dependencies.

Affected and fixed versions

  • OpenAI — GPT-5.5-Cyber (Daybreak / Trusted Access for Cyber)

Timeline

  • CyberGym benchmark (1,507 instances across 188 open-source projects, derived from OSS-Fuzz vulnerabilities) published (arXiv 2506.02548), later used to evaluate GPT-5.5-Cyber.
  • SEC-bench benchmark for evaluating LLM agents on real-world software security tasks published (arXiv 2506.11791); SEC-bench Pro later used as a GPT-5.5-Cyber evaluation.
  • OpenAI Codex Security plugin launched in research preview.
  • Axios reports OpenAI rolling out a more capable version of its cyber model with reduced safety refusals for approved defenders under Trusted Access for Cyber.
  • Trail of Bits publishes 'Introducing Patch the Planet,' detailing AI-assisted vulnerability research paired with human expert review and an initial five-day sprint surfacing hundreds of issues.
  • OpenAI expands its Daybreak cybersecurity platform; announces Patch the Planet initiative co-built with Trail of Bits and HackerOne, and the Codex Security plugin.
  • Threadlinqs Intelligence ingests the announcement as TL-2026-0909, an INFORMATIONAL industry/landscape signal documenting a dual-use AI capability (no CVE, no IOCs, no active exploitation).
  • GPT-5.5-Cyber full release covered by Cyber Security News: 85.6% CyberGym, 39.5% ExploitGym, 69.8% SEC-bench Pro; limited to verified, trusted defenders.

Sources cited

Previous and next in threat intel

Drafted by the Threadlinqs research pipeline from 10 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