SIAgentNews.com
← All news

OpenAI's cyber model reportedly found about 35 Apple flaws, including a Lockdown Mode bypass

A report says OpenAI's security-focused model helped expose Apple weaknesses and defeat the setting built for people most likely to be targeted.

· First reported by India Today on MSN

35Apple vulnerabilities reportedly found (approximate)

The short version

  • Researchers reportedly found around 35 flaws across iPhones, Macs, iPads, Apple Watches, Apple TVs and Vision Pro.
  • One exploit written with GPT-6 Cyber could reportedly strip away some Lockdown Mode protections.
  • Calif's CEO says Apple fixed many of the issues; Apple has not confirmed the full list.
  • Apple's security notes credit OpenAI Codex Security, not GPT-6 Cyber, for a handful of fixes.
  • GPT-6 Cyber is limited to approved customers under OpenAI's Daybreak Red programme.

Why it matters

Lockdown Mode is the setting people switch on when they fear they are being hunted, so a reported way around it matters more than an ordinary bug. The story also shows AI models now doing real vulnerability research, which both defenders and attackers will want. For everyday users, the practical lesson is that prompt updates matter.

What was reportedly found

As India Today on MSN reported, citing Forbes, OpenAI's newest security-focused model, GPT-6 Cyber, has been tied to roughly 35 security weaknesses across Apple's product line. A vulnerability is a flaw in software or hardware that an attacker might abuse. The list reportedly spans iPhones, Macs, iPads, Apple Watches, Apple TVs and Vision Pro headsets.

The work was reportedly done over several months by OpenAI together with the cybersecurity firm Calif. Some of the flaws were serious. A few could let an attacker take over an iPhone or Mac if the owner simply visited a booby-trapped website.

OpenAI launched GPT-6 Cyber on November 5 and described it as able to find and exploit security holes. The company rated it 'critical' for cybersecurity capability.

Why Lockdown Mode is the headline

Lockdown Mode is an optional Apple setting for people who could face sophisticated, targeted attacks. When it is on, the device switches off certain features and narrows the routes an attacker could use to get in.

According to the report, the researchers used GPT-6 Cyber to write an exploit, meaning code that takes advantage of a flaw, that could remove some of those protections. The report does not spell out how the bypass works or how complete it was, so it is best read as a sign of what the model could do rather than proof that the setting is broken.

Even so, the symbolism is strong. This is the feature meant to be the last line of defense, and an AI-assisted effort reportedly found a way to weaken it.

What Apple has and hasn't said

Calif's chief executive, Thai Duong, told Forbes that Apple patched many of the problems once it was notified. Apple, however, has not publicly confirmed the complete list of 35 and did not reply to Forbes' request for comment.

Some public evidence does exist. Apple's security notes dated October 26 for the iOS 27.2 and iPadOS 27.2 updates give credit to OpenAI Codex Security for three flaws in WebKit, the software Apple devices use to display web pages. OpenAI is also named alongside STAR Labs for a flaw in the kernel, the core layer of the operating system. Two of the WebKit problems could let an attacker run their own code through malicious web content. The macOS notes carry the same four credits.

Those credits are not a full accounting. They cover only a few items, and they name Codex Security rather than GPT-6 Cyber, so the two should not be treated as the same thing. The notes are also dated before the November 5 launch.

A model with a track record elsewhere

OpenAI says GPT-6 Cyber has been used beyond Apple. Its researchers reportedly used it to find 81 previously unknown high-severity flaws in Firefox, which Mozilla has begun fixing. In a separate exercise on the Linux kernel, the model helped produce working exploits for 948 vulnerabilities, 588 of which were new to the public.

OpenAI also says it passed 304 critical findings to Linux maintainers, with suggested patches attached. That detail matters because the stated aim of the tool is defensive: find the holes, prove they are real, and get fixes written before criminals get there.

For now, access is tightly controlled. Only approved customers can use the model, through a programme OpenAI calls Daybreak Red.

What to take from it

Treat this as a reported story with gaps. We do not have Apple's confirmation, a full list of flaws, or details on how much skilled human work surrounded the model. Those gaps limit how much anyone can conclude about what the AI did alone.

Still, the direction is clear. Security teams, whether on Apple's side or an attacker's, will increasingly use models to scan code, test ideas and draft exploits. The gap between a flaw existing and someone finding it may shrink, which makes installing updates quickly more important.

What to watch

  • Will Apple publicly confirm the full list of roughly 35 flaws and say which are fixed?
  • How complete was the Lockdown Mode bypass, and has Apple patched it?
  • Will OpenAI widen access to GPT-6 Cyber beyond its approved Daybreak Red customers?

Try it

Open Settings on your iPhone, iPad or Mac, check that automatic updates are on, and install any pending Apple update today.

Sources and links

First reported by India Today on MSN. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news