OpenAI's hacking AI found about 35 security flaws in iPhones and Macs
Forbes says GPT-6 Cyber turned up about 35 flaws, some able to take over a device; Apple's October updates already credit OpenAI with four fixes.
By The Graze · November 18, 2026 · 4 min read

Cybersecurity
Forbes reports that researchers using OpenAI's GPT-6 Cyber found about 35 security flaws across Apple's products, including some that could let an attacker take over an iPhone or a Mac through a malicious website. That "about 35" figure is Forbes's. OpenAI has not published a count of Apple bugs, and Apple's own security notes, released on 26 October, credit OpenAI on four fixed CVEs.
What Forbes reported, and what Apple has actually listed
The 35-flaw figure comes from Forbes, as set out in a 17 November report summarised by MLQ.ai, which checked the claim against Apple's notes and OpenAI's own page. Forbes said the work was done with Calif, that some of the flaws could allow a takeover through a malicious website, and that researchers also used the model to write an exploit that removes protections in Apple's Lockdown Mode. Calif told Forbes that Apple fixed many of the bugs once it was alerted. Apple had not responded to Forbes's request for comment, according to that account. Forbes did not publish a CVE-by-CVE list of the 35.
Apple's public credits are narrower, and they can be read in full. In the security content notes for iOS 27.2 and iPadOS 27.2, released on 26 October, Apple credits "OpenAI Codex Security" on three WebKit flaws: CVE-2026-112697, a use-after-free issue whose impact is that processing maliciously crafted web content may lead to arbitrary code execution; CVE-2026-112659, a type-confusion issue with the same stated impact; and CVE-2026-113903, a use-after-free issue whose impact is unexpected memory modification. On a fourth, kernel, flaw, CVE-2026-112698, Apple credits researchers at STAR Labs and OpenAI. The impact listed for that bug on iOS is that an app may be able to cause unexpected system termination.
The same four credits appear in the notes for macOS Golden Gate 27.2. The three WebKit fixes also shipped in Safari 27.2 for older Macs. Those are the OpenAI credits that can be counted, today, on Apple's own pages. They do not add up to 35, and they do not mention GPT-6 Cyber by name.
What OpenAI has said about the model
OpenAI's own announcement, dated 5 November, introduces GPT-6 Cyber as a cybersecurity model "trained to find and validate exploitable vulnerabilities, while staying within the scope that researchers set." The company classifies it as Critical in cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber. It is initially available through the API and Codex to a select group of approved Daybreak Red customers.
The official page does not mention Apple, or a total of about 35 Apple flaws. It does give verified figures for other campaigns. OpenAI said its researchers used GPT-6 Cyber to identify 81 previously unknown high-severity vulnerabilities in Firefox, and that Mozilla has begun landing fixes. It also said the model reproduced 1,724 Linux kernel vulnerabilities and developed working exploits for 948 of them, more than five times the 166 exploited by GPT-5.6 Sol using the same research harness, and that 304 critical findings were disclosed to maintainers with candidate patches.
Calif is named on that page, not as the author of an Apple study, but as one of the security teams, with Cisco and CodeRabbit, already using the model for vulnerability research and code review.
| Item | What can be verified | Source |
|---|---|---|
| Forbes's Apple tally | About 35 vulnerabilities across Apple hardware and software, including some that could allow a takeover via a malicious website | Forbes, 17 November 2026, as reported by MLQ.ai. OpenAI has not published this number. Forbes published no CVE list. |
| Apple credits to OpenAI, 26 October 2026 | Four CVEs: WebKit CVE-2026-112697 and CVE-2026-112659 (arbitrary code execution); WebKit CVE-2026-113903 (unexpected memory modification); kernel CVE-2026-112698, with STAR Labs (unexpected system termination on iOS) | Apple's notes for iOS 27.2, iPadOS 27.2, macOS Golden Gate 27.2 and Safari 27.2 |
| GPT-6 Cyber launch | 5 November 2026; rated Critical; limited to approved Daybreak Red customers | OpenAI |
| Other official campaign figures | 81 previously unknown high-severity Firefox vulnerabilities; working exploits for 948 Linux kernel flaws | OpenAI. Not Apple figures. |
Compiled by The Graze from OpenAI's 5 November 2026 announcement of GPT-6 Cyber, Apple's 26 October 2026 security notes for iOS 27.2, iPadOS 27.2, macOS Golden Gate 27.2 and Safari 27.2, and MLQ.ai's 17 November account of the Forbes report. The "about 35" tally is Forbes's.
Our take
There are two numbers in this story, and they should not be sanded into one. "About 35" is a magazine figure, attributed to work with Calif, with no public list of bugs behind it. Four is the number of Apple CVEs that, as of the 26 October notes, name OpenAI as a credited finder. The second number is smaller, official and checkable. The first may well include lower-severity items, bugs fixed without a public credit, or findings that have not yet appeared in a security note. It may also be roughly right. From the outside, there is no way to know, because neither OpenAI nor Apple has published a tally that matches it.
What can be said, without the 35, is still substantial. A restricted OpenAI model, out only since 5 November and already classed Critical by the company that made it, is being used to find and prove real bugs in widely used software. Apple's notes show that some of that work, or work to which OpenAI's name is attached, reached iPhones, iPads and Macs, including two WebKit flaws that Apple itself says can lead to arbitrary code execution from a malicious web page. That is the part of the Forbes account that has a public footprint. The practical step for users is the ordinary one: install the October updates, and the ones that follow. The practical step for everyone else is to keep the two figures separate until someone publishes the list.
Sources
- OpenAI, GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents, 5 November 2026
- Apple Support, About the security content of iOS 27.2 and iPadOS 27.2, 26 October 2026
- Apple Support, About the security content of macOS Golden Gate 27.2, 26 October 2026
- Apple Support, About the security content of Safari 27.2, 26 October 2026
- MLQ.ai, OpenAI's GPT-6 Cyber found about 35 flaws in Apple's iPhone and Mac software, Forbes reports, 17 November 2026







