Apple is facing an overload of bug reports due to AI significantly lowering the barrier to vulnerability discovery. In October 2025, Apple increased its bug bounty to $5 million, but AI tools have enabled amateur researchers to bulk-scan code and submit large volumes of false vulnerabilities fueled by AI hallucinations. On August 2, Apple implemented submission limits and a 30-day cooling-off period in response. Additionally, Apple’s five-year-developed MIE memory safety feature was breached in just five days by security firm Calif using the Claude model, forming a complete privilege escalation chain. CVE registrations are projected to reach 66,000 in 2026—46% higher than estimated. Google, Nextcloud, GitHub, and others have already taken steps to restrict AI-generated vulnerability reports. On July 27, Apple released macOS Tahoe 26.6 security update, patching 194 vulnerabilities, and for the first time formally acknowledged AI tools such as Anthropic Claude and OpenAI Codex Security.Author and source: WeChat public account "Quantum Bit"
Apple has been forced to implement a "bug submission cooling-off period."
Apple has finally had enough and implemented a cooling-off period for its bug bounty program.
Urgent, too many bugs to review… haha, never mind! I’m not looking anymore 😈.The bug bounty program, introduced in 2016, was originally designed to invite security researchers to submit vulnerabilities in Apple's software, hardware, and services, with cash rewards provided upon manual review and confirmation of valid reports.
The bounty pool continued to grow, and in October 2025, Apple announced a major upgrade to its bounty program: researchers who discover the most severe and complex threats can receive $5 million (approximately 33.75 million RMB).
However, currently, AI has significantly lowered the barrier to finding vulnerabilities, with many amateur users using ChatGPT to scan code in bulk and submit reports en masse, many of which contain AI-generated hallucinations of false vulnerabilities.
This has completely overwhelmed the entire Apple security team.
On August 2, Apple set a limit on the number of vulnerability submissions and a 30-day cooling-off period on its internal security portal.
Apple's strongest defense system has been easily bypassed by AI
In September 2025, Apple unveiled a security feature called Memory Integrity Execution (MIE) during the iPhone 17 launch event.
This technology, which took five years to develop, leverages the hardware capabilities of Apple’s custom chips and deep integration with the operating system to provide unparalleled memory security for devices.
Apple calls this feature:
The most significantupgradein the history of consumer-grade operating system memory security.
Diagram of MIE technical principles: How MIE prevents memory attacks
However, eight months later, the cybersecurity industry experienced a collective awakening.
In May 2026, security research firm Calif disclosed the first publicly known exploit for Apple M5 chip macOS. Its employees chained together two macOS vulnerabilities into a complete local privilege escalation chain, bypassing MIE protections, using Claude’s Mythos Preview model.
And the team (just three people) went from “we found something interesting” to “we achieved a working root shell on Apple’s most secure consumer hardware” in just five days.
The severity of the vulnerability ultimately led the two researchers from Calif to drive personally to Apple’s headquarters to hand in their 55-page report, to ensure it wouldn’t be lost among numerous other submissions.
Mythos Preview, launched by Claude in April this year. The official statement claims it has achieved a qualitative breakthrough in security and programming, and has autonomously discovered thousands of critical vulnerabilities in every major operating system and mainstream web browser. However, due to the model's "excessively strong offensive capabilities, which could cause a major disaster if made public," the API has been opened only to 40 key partners.
The rapidly advancing AI technology is disrupting the entire vulnerability disclosure system.
It is predicted that the number of CVEs (publicly disclosed cybersecurity vulnerabilities) registered in 2026 will reach 66,000, 46% higher than the original estimate.
However, as Apple has discovered, these AI-generated vulnerability reports contain a significant amount of noise and AI hallucinations, making their review more labor-intensive.
This is also a pain point across the entire security industry.
This is a fairly difficult period in the industry, with maintenance teams and suppliers already overwhelmed by a large number of bugs.The creator of cURL stated that in the first three weeks of 2026, he received 20 vulnerability reports, none of which were genuine security vulnerabilities. Google announced in March that it would no longer accept AI-generated vulnerability reports; the open-source cloud platform Nextcloud paused its bug bounty program in April; and GitHub significantly reduced the payout amounts for public vulnerability bounties in July, establishing a VIP channel accessible only to invited researchers.
The congestion in this submission channel has already led to real security consequences and may turn the security process itself into a security issue.
For example, the Italian security startup Bynario used ChatGPT to discover over 50 vulnerabilities in macOS within three weeks, including one that could fully compromise macOS.
However, when the team submitted this vulnerability, which they claimed could be sold on the black market for $100,000 to $200,000, they found that Apple had already blocked the submission portal.
Apple's security update thanks AI for the first time
The good news is: defenders can also access the same AI tools.
The bad news is: defenders must keep the entire IT environment running smoothly, test compatibility, coordinate release windows, and ensure updates don’t cause production systems to go down. Attackers? They only need to find one entry point.
On July 27, 2026, Apple released the macOS Tahoe 26.6 security update, which fixed 194 distinct security vulnerabilities.
This release also marks Apple's first official acknowledgment of AI in its security patches.
Discover with Claude
Among them, Anthropic's Claude and OpenAI's Codex Security each received three acknowledgments, NVIDIA's AI Red Team received two acknowledgments, and Z.ai's GLM model received one acknowledgment.
(Although three weeks ago, Apple was suing OpenAI in court for stealing trade secrets.)
Apple has publicly stated that AI tools have accelerated the release of our security updates. Looking at the security update data since the release of macOS Tahoe, we can see the outline of this acceleration trend:
AI tool attribution appeared in version 26.5; version 26.5.2, an irregular minor update, added fixes for 38 additional CVEs; version 26.6, released a month later, broke the record again with 155.
However, it is uncertain whether this high-frequency release rhythm itself will become a new security variable.
After all, Apple has always been known for its strict control over release schedules, and every system update undergoes internal testing, compatibility verification, and phased rollouts before being pushed to billions of devices.
This suggests that accelerating the release was not a hasty decision, but rather an assessment by Apple’s security team that the risk of waiting until the next regular update window to fix these vulnerabilities was unacceptable.
The vulnerability disclosure cycle is shifting: as AI compresses the vulnerability discovery timeline to just a few days, the pace of monthly security updates may become a prolonged period of exposure.