OpenAI’s GPT-6 Cyber Found ~35 Apple Flaws; Apple Credits 5

OpenAI and cybersecurity company Calif have used OpenAI’s GPT-6 Cyber model over recent months to uncover “some 35 weaknesses of varying severity across Apple hardware and software”, Forbes reported on 17 November. Forbes says some “could allow a hacker to take over a Mac or an iPhone via a malicious website”.

Our number: across 27 of Apple’s own security-content pages for releases dated 14 September to 13 November 2026, we count 608 CVE IDs. Apple’s credit lines name OpenAI on 5 of them and Calif researchers on 17, so 22 name OpenAI or Calif, by our count. Seven of the 17 Calif credits read “in collaboration with Claude and Anthropic Research”.

Business Pill · Red Teaming

An evergreen Business Engineer explainer on red teaming: attacking your own systems to find weaknesses first. It teaches the general idea only and does not discuss Apple, OpenAI or this report.

The key insight: Forbes reports that OpenAI and Calif used GPT-6 Cyber to find some 35 Apple vulnerabilities; Apple’s own pages since 14 September name OpenAI on 5 CVEs and Calif on 17, by our count, and credit 7 of those Calif fixes in collaboration with Anthropic’s Claude.

What Forbes Reported

Forbes says GPT-6 Cyber “has found security flaws in nearly everything the company makes, including iPads, Apple Watches, AppleTVs and Vision Pro headsets”. It adds that researchers “were also able to use GPT-6 Cyber to code an exploit capable of removing the protections provided by Apple’s Lockdown mode”.

Forbes describes Calif as “a close partner of Sam Altman’s AI giant”. Calif CEO Thai Duong told Forbes: “I think things will become worse before they become better”, and that “between now and then, I think there will be more bugs, more exploits, and not just for the iPhone, but for the internet in general”.

On the fixes, Forbes reports that “Apple has not yet responded to requests for comment; Calif told Forbes the company fixed many of the vulnerabilities when it was alerted to them”. It adds: “OpenAI confirmed that its latest model had found vulnerabilities for which Apple credited it, but did not comment further”.

Forbes reports some 35 Apple vulnerabilities found by OpenAI and Calif with GPT-6 Cyber. Across 27 of Apple's
Forbes reports some 35 Apple vulnerabilities found by OpenAI and Calif with GPT-6 Cyber. Across 27 of Apple’s security-content pages for releases dated 14 September to 13 November 2026 (608 CVEs), Apple’s credit lines name OpenAI or Calif on 22 CVEs: Calif on 17, OpenAI on 5 (3 as OpenAI Codex Security); 7 of the Calif credits read ‘in collaboration with Claude and Anthropic Research’. Counts are ours; credit lines do not name the OpenAI model, and Forbes did not list the 35.

What Apple’s Own Pages Show

Apple lists its fixes on its security releases page. We read the security-content pages for 27 of the 30 most recent entries, dated 14 September to 13 November 2026; three pages (watchOS 26.6.1, tvOS 27.2 and visionOS 27.2) could not be read. Across them we count 608 unique CVE IDs.

Five of those CVEs carry a credit line that names OpenAI, three of them as “OpenAI Codex Security – Amy Burnett”. Seventeen name researchers of Calif, written as Calif.io or (Calif). That is 22 CVEs in all, against the “some 35” Forbes reports, by our count.

The credit lines do not name the model used, except where Apple writes “in collaboration with Claude and Anthropic Research”, which appears on 7 of the 17 Calif-credited CVEs, each naming Bruce Dang of Calif.io or Calif.io. So the same firm Forbes describes as a close OpenAI partner is also credited by Apple alongside Anthropic’s Claude, on Apple’s pages.

How Fast and How Costly

Forbes cites a test by XBOW, an AI security startup that “tasked GPT-6 Cyber with finding security issues in 65 open-source web apps with known vulnerabilities”. Forbes reports the model “is able to find and exploit a weakness twice as fast as GPT-6.1 Sol”, and that “$1 spent on Sol, an older OpenAI model, would likely uncover the same number and kinds of exploits as $17 spent on GPT-6”.

Forbes also says that when OpenAI launched GPT-6 Cyber on 5 November, “the company said it had helped identify over 2,000 bugs in the Linux open source operating system and developed successful hacks for 948 of them”. We covered the GPT-6 Cyber launch and its comparison with Anthropic’s Mythos.

The Business Lever

The lever is offensive security as a product line for the labs. OpenAI President Greg Brockman told The Information there is no question cybersecurity will be a bigger business than coding, and OpenAI says NATO used up to 300 billion of its tokens a month for cyber defence. Apple’s credit lines show both OpenAI and Anthropic’s Claude attached to fixes in the same release cycle.

Forbes reports that XBOW founder Oege de Moor and Duong “stress that AI is making hacking considerably more efficient, not making human ingenuity obsolete”, and that “A human hacker has to supply the methodology and verify that the finding is real”.

The Structural Read

As we read Apple’s pages, the public record lags the reported total: 22 credits name OpenAI or Calif against the some 35 Forbes reports, by our count, and the lines do not say which model found each bug.

As we read the credit lines, the same research firm works with both labs: Calif is described by Forbes as a close OpenAI partner, and Apple credits 7 Calif fixes in collaboration with Claude and Anthropic Research.

Thai Duong, Calif CEO, to Forbes

“I think things will become worse before they become better”

Three Implications

FOR APPLE USERS Forbes reports Calif told it Apple fixed many of the vulnerabilities when alerted, and that Duong still recommends Lockdown Mode even though his team found a way around some of its protections.

FOR OPENAI Forbes reports OpenAI confirmed its latest model had found vulnerabilities for which Apple credited it; Apple’s pages name OpenAI on 5 CVEs since 14 September, three as OpenAI Codex Security, by our count.

FOR SECURITY BUYERS Forbes cites XBOW finding GPT-6 Cyber twice as fast as GPT-6.1 Sol on 65 web apps, but $1 of Sol would likely uncover the same exploits as $17 of GPT-6, as reported.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Agentic Harness War.

The framework says: “Value capture migrates up from the model to the harness.”

As we read it, Forbes’ account fits that pattern: XBOW and Duong say a human hacker still supplies the methodology and verifies findings, as reported, so the security firms wrapping the model do part of the work. The framework predates this story and does not discuss Apple or GPT-6 Cyber.

What Is Not Established

Forbes does not list the 35 vulnerabilities or their CVE IDs, and Apple had not responded to Forbes. We could not match Forbes’ “some 35” to Apple’s pages: our count of 22 covers only credits that name OpenAI or Calif on 27 pages since 14 September, and some fixes may be credited differently, listed on the three pages we could not read, or not yet published. The credit lines do not say which OpenAI model found each bug.

The viral posts on X said a “New OpenAI model” found the flaws. Forbes names the model as GPT-6 Cyber, launched on 5 November, not the GPT-6.2 Astra and Sol models OpenAI released on 17 November. We did not contact OpenAI, Apple or Calif.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

The Agentic Harness War

A Business Engineer analysis of where value pools as AI agents move from the model to the harness around it.

Read the Map of AI →

The Bottom Line

Forbes reports that OpenAI and Calif used GPT-6 Cyber to find some 35 Apple vulnerabilities, including some that could let a hacker take over an iPhone or Mac via a malicious website. Apple’s own pages since 14 September name OpenAI on 5 CVEs and Calif on 17, by our count, and credit 7 of the Calif fixes in collaboration with Anthropic’s Claude.

95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. The figure of some 35 vulnerabilities, the Lockdown Mode exploit, the quotes from Calif and XBOW, the XBOW speed and cost figures and the statements from OpenAI and Apple are as reported by Forbes on 17 November 2026; we read the article as served. The CVE counts are our count of the credit lines on 27 of Apple’s security-content pages for releases dated 14 September to 13 November 2026, read 17 November. We did not contact OpenAI, Apple or Calif.

Sources: Forbes, OpenAI’s New Hacking Model Found Dozens Of Holes In Apple Tech, 17 Nov 2026 · Apple, Apple security releases and security-content pages, read 17 Nov 2026 · Thomas Brewster on X, 17 Nov 2026

Scroll to Top
FourWeekMBA