OpenAI's GPT-6 Cyber found about 35 Apple flaws, Forbes reports

Security firm Calif used OpenAI's restricted hacking model to find bugs across Apple devices, including iPhones and Macs, and to build a Lockdown Mode bypass.

Illustration of a smartphone and a laptop under a magnifying glass, with orange warning markers on their circuitry and a robotic arm holding a patch over one of them. Image: AgenticFeed illustration

Researchers working with OpenAI's GPT-6 Cyber model found about 35 security vulnerabilities across Apple hardware and software, including flaws that could let an attacker take over an iPhone or Mac through a malicious website, Forbes reported on Tuesday.

The work was done by OpenAI and Calif, a security research firm Forbes described as a close partner of OpenAI. According to the report, the flaws vary in severity and span iPhones, Macs, iPads, Apple Watches, Apple TVs and Vision Pro headsets.

Researchers also used the model to write an exploit that removes the protections of Lockdown Mode, the hardened setting Apple offers people at high risk of targeted attacks, Forbes reported.

What Apple has credited

Calif told Forbes that Apple fixed many of the bugs after it was alerted. OpenAI confirmed to the outlet that its latest model found vulnerabilities Apple credited it for but did not comment further, and Apple did not respond to a request for comment, according to Forbes.

Apple's public security notes list a smaller set.

According to its notes for iOS 27.2 and iPadOS 27.2, Apple credits "OpenAI Codex Security" on three WebKit flaws and lists OpenAI alongside STAR Labs on a kernel bug.

Apple said two of the WebKit issues could lead to arbitrary code execution when processing malicious web content. The same credits appear in Apple's notes for macOS Golden Gate 27.2. Forbes did not publish a full list of the 35 findings.

A model rated Critical

OpenAI launched GPT-6 Cyber on Nov. 5 and rates it "Critical" for cybersecurity under its Preparedness Framework, up from the "High" rating of GPT-5.6 Cyber, The Verge reported. Under that framework, a Critical rating means a model "can identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention," according to The Verge.

OpenAI said the model helped find 81 previously unknown vulnerabilities in Firefox, The Verge reported. Access is limited to select members of OpenAI's Daybreak Red program for authorized defensive work, with a wider rollout to all Daybreak Red customers planned in the coming weeks, according to the same report.

Speed and cost

Forbes also cited testing by AI security startup XBOW across 65 open-source web apps with known bugs. XBOW found that GPT-6 Cyber discovered and exploited weaknesses twice as fast as OpenAI's older GPT-6.1 Sol model, but estimated that $1 spent on Sol would likely find the same number and types of exploits as $17 spent on GPT-6 Cyber, Forbes reported.

Calif CEO Thai Duong and XBOW CEO Oege de Moor both said a human hacker still has to supply the methodology and verify that findings are real, according to Forbes.

Worse before better

Duong told Forbes the new model is a step up in helping write exploits at speed, which he described as a benefit for defenders and a liability in the hands of malicious hackers.

"I think things will become worse before they become better," Duong said. He added that "maybe in six months or a year, we can finally eliminate a lot of bugs and models can generate patches for the bugs autonomously," but said "we're not there yet."

Calif drew attention last month when it used OpenAI models to find flaws in PHP, the web programming language, which led many companies to take sites offline while fixes rolled out, Forbes reported. Apple's credited fixes so far ship in iOS 27.2, macOS Golden Gate 27.2 and later updates, according to its security notes.

Source: Forbes, “OpenAI's New Hacking Model Found Dozens Of Holes In Apple Tech”

Additional sources: Apple, “About the security content of iOS 27.2 and iPadOS 27.2”; The Verge, “OpenAI says GPT-6 Cyber is better at finding vulnerabilities without going out of bounds”

Get the AgenticFeed newsletter

A free, regular roundup of the agent stories that matter. No spam, and you can unsubscribe at any time.

Want to reach readers following AI agents? Advertise with AgenticFeed

Next event · 19 Nov, 2026

MCP Dev Summit Nairobi

The Linux Foundation’s Nov. 19 and 20 summit brings Model Context Protocol developers to Kenya, with keynotes from OpenAI, GitHub and the Gates Foundation.

The AgenticFeed newsletter

A free, regular roundup of the agent stories that matter. No spam.