
GPT-6 Cyber found about 35 flaws in Apple software
Forbes says GPT-6 Cyber and Calif found about 35 Apple vulnerabilities. Apple’s Oct. 26 notes credit OpenAI on four CVEs, including three WebKit flaws and one kernel issue.
- By
- Rebecca Stone
- Filed
- Channel
- AI & Compute
- Read
- 4 min read
By Rebecca Stone
OpenAI’s GPT-6 Cyber found about 35 security vulnerabilities across Apple hardware and software, including flaws that could let attackers take control of an iPhone or Mac through a malicious website, Forbes reported on Tuesday, Nov. 17.
The researchers worked with security firm Calif, which told Forbes that Apple fixed many of the bugs after receiving an alert. They also used GPT-6 Cyber to write an exploit that removes the protections of Lockdown Mode, Apple’s hardened setting for people at high risk of targeted attacks.
That claim is broader than Apple’s public record. The company has not published a list matching the roughly 35 findings, and Forbes did not provide a CVE-by-CVE accounting. OpenAI confirmed to Forbes that its latest model found vulnerabilities Apple credited to it; OpenAI declined to comment further, and Apple did not respond to Forbes’s request for comment.
What has Apple confirmed?
Apple’s Oct. 26 security notes for iOS 27.2 and iPadOS 27.2 credit “OpenAI Codex Security” with three WebKit flaws:
- CVE-2026-112697 — WebKit.
- CVE-2026-112659 — WebKit.
- CVE-2026-113903 — WebKit.
They also credit OpenAI alongside STAR Labs on one kernel flaw:
- CVE-2026-112698 — kernel bug, jointly credited.
Two of the WebKit bugs could lead to “arbitrary code execution when processing malicious web content,” Apple said.
The same four credits appear in the macOS Golden Gate 27.2 notes. The three WebKit fixes also shipped in Safari 27.2 for older Macs.
The October notes separately credit several researchers from Calif.io on bugs in WebKit, CoreAudio, CoreGraphics and video-encoder components. Forbes did not publish a CVE-by-CVE list of all 35.
The public record therefore supports two separate conclusions. First, Apple’s release notes tie OpenAI and STAR Labs to four fixed flaws, and OpenAI confirms its model found bugs in that credited set. Second, the much larger estimate depends on Forbes’s reporting because neither Apple nor Forbes has published the complete set.
Calif’s role spans the research and disclosure chain. The firm told Forbes it alerted Apple and that many bugs were fixed, while Apple’s notes separately name Calif.io researchers on other components. Forbes’s account does not map every finding in the roughly 35-item set to a CVE and discoverer.
What can GPT-6 Cyber do, and who can use it?
OpenAI announced GPT-6 Cyber on Nov. 5 and rates it “Critical” for cybersecurity under its Preparedness Framework, a step above the “High” rating of its previous cyber model. The company calls it a restricted hacking-and-defense model and limits access to approved customers in its Daybreak Red program.
In pre-launch testing, OpenAI said the model found 81 previously unknown high-severity Firefox vulnerabilities and built working privilege-escalation exploits for 948 distinct Linux kernel flaws.
Those remain OpenAI’s own test results. They do not replace the public record of what shipped in Apple’s October releases, where four OpenAI-linked CVEs appear.
How does this change the AI-security contest?
Earlier on Tuesday, OpenAI said NATO used up to 300 billion tokens a month of OpenAI’s cyber-defense models. Separately, Google has warned that attackers are using AI agents to find and exploit flaws with no human in the loop.
Together, those disclosures put GPT-6 Cyber’s defensive value alongside its offensive reach. The Apple work reportedly used one system for both vulnerability discovery and an exploit against Lockdown Mode, while NATO’s reported token consumption shows how extensively OpenAI’s cyber-defense models were already in use.
What should iPhone and Mac users install now?
The fixes Apple has credited so far ship in iOS 27.2, iPadOS 27.2, macOS Golden Gate 27.2 and later updates. Older Mac users can get the three WebKit fixes through Safari 27.2.
For iPhone and Mac users, the immediate step is therefore straightforward: install the credited fixes rather than wait for a complete list of the roughly 35 findings.
Going forward, the competitive stakes are remediation speed, not discovery alone. GPT-6 Cyber has paired a roughly 35-vulnerability Apple claim with a reported Lockdown Mode exploit, while Apple’s public notes confirm four CVEs; the commercial test is whether the remaining findings produce faster patches before attackers can turn them into working exploits.
Source: mlq.ai
More from Rebecca Stone
Show full bio
Correspondent covering technology and product news in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.
1,092 articles
Related articles
openai-s-gpt-6-cyber-reproduced-1-724-kernel-bugs-as-agent-tools-launch
OpenAI's GPT-6 Cyber Reproduced 1,724 Kernel Bugs as Agent Tools Launch
ivanti-and-zoom-patch-critical-flaws-as-citrix-and-asus-ship-fixes
Ivanti and Zoom Patch Critical Flaws as Citrix and Asus Ship Fixes
java4shell-openjdk-flaw-enables-unauthenticated-code-execution
Java4Shell: OpenJDK Flaw Enables Unauthenticated Code Execution


