Apple’s security researchers have a new headache, and it isn’t a rogue nation-state. It’s the sheer volume of vulnerability reports pouring in from people wielding large language models — a phenomenon the industry has cheerfully dubbed AI slop.
Confirmed to The Financial Times, Apple has reworked its bug bounty program to keep review teams above water. As of June 2026, the company introduced a cap on open submissions and a 30-day cool-off period for reports sent through its internal security portal. Researchers who hit the ceiling must formally request an increased quota before they can file more.
The trigger is obvious to anyone watching the space. Modern LLMs can now find, chain and exploit vulnerabilities at a pace no human triage team can match. The result is an avalanche — some of it genuinely brilliant, much of it noise — that clogs the pipeline and buries the reports that actually matter.
And the good stuff is very good. Apple recently accelerated its security patches in direct response to these tools, shipping fixes in iOS 26.5.2 and its sibling releases that had originally been earmarked for the version 26.6 updates. In the accompanying security notes, Apple credited researchers who leaned on AI models from OpenAI, Anthropic, Z.ai and others for surfacing several flaws.
One standout: the team at Calif.io said in May that it used Anthropic’s Mythos Preview model to build a working macOS kernel memory-corruption exploit on M5 silicon in just five days. That is the kind of timeline that keeps defenders awake at night.
Apple isn’t alone in tightening the taps. Days before the FT report, GitHub rolled out a tiered system for its own bounty program, designed to filter out slop while giving verified security researchers a clearer path.
The policy shift has real casualties, though. The FT profiled Bynario, a seven-person cybersecurity start-up using recent AI models to hunt bugs. After reporting five vulnerabilities to Apple this year — and eight last year, one of which was patched in a November software update — the firm found its submissions blocked. Following the FT’s reporting, Apple reopened contact with Bynario and began reviewing its findings, including a privilege-escalation exploit chain that could hand an attacker full control of a Mac.
In its statement, Apple framed the change as pragmatic housekeeping: “With the growing volume of AI-generated security submissions across the industry, we recently adjusted the number of new reports a researcher can have open at once.” The company added that researchers “can easily request an increase to that limit at any time to ensure critical reports reach our security teams.”
The tension is clear. AI is now one of the most productive vulnerability hunters on the planet — and one of the noisiest. Apple’s cap is a bet that quality can be protected without drowning the people who read it.
